The research is basically not neuroscience or even neurophysiology, but mathematical modeling and speculation. Here is the key passage from the abstract:
The main point is that short-term memory processes are supposed to stop small-error correction when short-term memory is the main basis for learning as supposedly in "blocked" practice, and this stooping happens sooner the quicker the short-term memory learns the movement. This is supposed to be a bad thing for long-term memory which needs more time to experience the small-error corrections. The idea is that both forms of working memory rely upon the small-error corrections to build up the skill / movement memory, and if the "blocked" practice uses only or mostly the short-term working memory, the long-term memory of the skill is not stable because deprived of long-enough small-error corrections.
"These counterintuitive results were predicted by a computational model of motor memory that contains a common fast process and multiple slow processes, which are competitively updated by motor errors. In blocked schedules, the fast process quickly improved performance, therefore reducing error-driven update of the slow processes and thus poor long-term retention. In random schedules, interferences in the fast process led to slower change in performance, therefore increasing error-driven update of slow processes and thus good long-term retention. Increased forgetting rates in the fast process, as would be expected in individuals with visuospatial working memory deficits, led to small updates of the fast process during blocked schedules and thus better long-term retention."
That's not neuroscience, but basic 1970s-1980s style motor sports science, built upon suppositions and concepts that may or may not reflect neuroscience reality.
The standard motor sports science of the "contextual interference effect" is reviewed here: "A review of the contextualinterferenceeffect in motor skill acquisition," Richard A. Magill, and Kellie G. Hall, Human Movement Science, Volume 9, Issues 35, September 1990, Pages 241289:
"The contextualinterferenceeffect is a learning phenomenon where interference during practice is beneficial to skill learning. That is, higher levels of contextualinterference lead to poorer practice performance than lower levels while yielding superior retention and transfer performance. This rather counterintuitive effect, first demonstrated by Battig (1966) for verbal materials and later shown to be relevant to motor skill learning by Shea and Morgan (1979), has led to a considerable amount of research. In this article, we review the motor skills literature related to this effect by focusing on two research directions. First, evidence related to the generalizability of the effect is considered to identify possible parameters of the contextualinterferenceeffect. It is apparent that this effect is not applicable for learning all types of tasks and for all types of learners. Thus, task and individual characteristics are given particular consideration. Second, the explanations offered to establish what learning processes account for this effect are considered. Here, the different views proposed to explain the contextualinterference are discussed along with research addressing this issue. Also, the relationship between the contextualinterferenceeffect and the practice variability hypothesis of Schmidt's schema theory is explored. Finally, future research directions are suggested with the goal of providing guidelines for research to enhance our present knowledge of the contextualinterferenceeffect and its relationship to motor skill learning."
Motor sports science folks are pretty confused about all this, primarily because they deal with concepts about the brain and not with the actual mechanisms of the brain. here is a typical discussion showing the confusion and the straining to incorporate modern neuroscience into the old bottles of the motor sports concepts from the pre-1990s:
"In the study of learning and memory, varied practice (also known as variable practice or mixed practice) refers to the use of a training schedule that includes frequent changes of task so that the performer is constantly confronting novel instantiations of the to-be-learned information.
The varied practice approach focuses on the distribution of practice in time, the organization of activities to be practiced (blocked vs. random), and the interleaving of information or content to highlight distinctions that facilitate learning. For example, a varied practice approach to learning to shoot a basketball might involve a sequence of ten mid-range jump shots, followed by ten lay-ups, followed by ten free-throws, followed by ten three-pointers, with the entire cycle repeating ten times. This contrasts with traditional approaches in which the learner is encouraged to focus on mastering a particular aspect or subset of the relevant information before moving on to new problems (e.g., focusing on free-throws before moving to three pointers). With varied practice, the learner is exposed to multiple versions of the problem even early in training.
In many learning domains, varied practice has been shown to enhance the retention, generalization and application of acquired skills. There are many potential sources of the observed advantages. First, greater diversity of the tasks may also allow learner to extract the most relevant, task-invariant information. Any given practice trial contains both task-relevant and task-irrelevant information. By mixing up the trials, task-irrelevant information will be less consistent, allowing the learner to strip away the spurious associations. Task-relevant information should be constant regardless of the particulars of individual trial. Second, varied practice creates conditions that are likely to encourage elaborative rehearsal (see Craik & Tulving, 1975). Elaborative rehearsal is a means by which the learner forms multiple associations with the to-be-learned material, so that it can be recalled using a variety of cues. Cognitive psychologists generally regard elaborative rehearsal as one of the most effective means of acquiring new information, and its basic logic to study the material from a range of perspectives in order to former richer links with preexisting knowledge is completely consistent with the varied practice approach. Finally, because learners are frequently changing tasks, practice may seem less repetitive, potentially minimizing boredom and increasing the level of engagement during practice.
The theoretical underpinnings of the varied practice approach stem primarily from a behavioral phenomenon discussed in the skill acquisition literature called contextual interference (Shea & Morgan, 1979). Contextual interference refers to a learning benefit observed when the items to be learned are randomly intermixed across training blocks rather than repeated in blocks (for a review, see Magill and Hall, 1990). That is, when identical items are blocked together during training, post-training performance is worse than when different items are intermixed. Although primarily studied with motor skill learning task, contextual interference was originally reported in a verbal paired associates task (Battig, 1966, 1972) and has been observed in other nonmotor tasks (e.g., Carlson et al., 1989). Intriguingly, the benefits of mixed-item blocks are apparent only some period of time after practice, indicating that the effects are primarily long-term.
The source of contextual interference is not well understood. At present, it is primarily an empirical phenomenon. Most accounts assume that it emerges because blocked practice is not sufficiently demanding to produce optimal effort or attention. Consistent with this view, contextual interference is reduced or eliminated with more complex tasks (see Wulf & Shea, 2002). Across the various accounts of this complexity effect, the dominant theme is that as complexity increases, learners benefit more from the opportunity to repeat and refine their responses on successive trials. There is also debate regarding whether children show the effects of contextual interference as adult learners. Some studies suggest children show normal contextual interference (e.g., Edwards et al., 1986) whereas others show no effect in children (e.g., Del Rey et al., 1983). Moreover, as with adults, more difficult or complex tasks show less contextual interference (Magill & Hall, 1990). Thus, the absence of contextual interference in some studies with children may simply be another manifestation of the task complexity effect.
Contextual interference bears an intriguing similarity to a phenomenon observed during the training of neural networks called catastrophic interference (McCloskey & Cohen, 1989). Catastrophic interference occurs when a network is trained to criteria on one set of mappings, and then switched to a new set, at which point it loses access to the initial mappings. In a sense, rather than forming a set of connections that would preserve the knowledge it acquired in the first task, the network optimizes its performance completely to the new task. The solution to this problem is simply to interleave the training sets so that the network is forced to optimize its behavior in a way that is sensitive to both of the tasks and their statistics. While this was not thought to characterize human learning (a supposed weakness of this approach), recent evidence suggests that human learning also exhibits this principle under the right conditions (Mirman & Spivey, 2001).
Neuroscience techniques have recently been applied to investigating the effects of varied practice. Because these effects are typically only observed after a considerable delay, these studies have focused on the neural changes occurring during the consolidation period. For example, Kantak and colleagues (2010) used repetitive transcranial magnetic stimulation (rTMS) immediately after varied practice of a motor task and, in a separate group of participants, after traditional constant practice of the same task. When the rTMS was applied over the prefrontal cortex, performance 24 hours after training was much worse than immediately after training. In contrast, when the rTMS was applied over primary motor cortex, performance 24 hours after training improved similarly to a control group that received no rTMS. Thus, disruption of neural activation in the prefrontal cortex after varied practice appears to eliminate its benefits. In contrast, when a constant practice schedule was used, all groups performed worse 24 hours after training than immediately after training, and the group receiving rTMS over the primary motor cortex performed much worse than either the prefrontal or control groups. In sum, the findings suggest that varied and constant practice engage distinct sets of neural regions, the former involving the prefrontal cortex and the latter involving primary motor cortex."
So, the neuroscience is pretty sketchy but suggests that varied / random practice learning occurs mostly in the prefrontal cortex, whereas blocked / constant practice learning occurs mostly in the primary motor cortex. This neuroscience is pretty incompatible with the Old-bottles language in the study you bring to the Flatstick Forum. No surprise there -- mathematical modeling a la "neural networks" is not looking at the actual processes of the brain.
Once again, motor sports people cling desperately to the old-bottle concepts. Not to be trusted.
Unfortunately, one can expect a steady stream of this sort of research.
Putting Coach and Theorist