My understanding of motor skills learning suggests that the question in the form as you pose it glosses over some complications that have to be respected. By that I mean the a) or b) form of the question is not how I would discuss the subject.
The old-school motor learning theory was not much more complex than the notion that repetition builds muscle memory, and that learning a new motion required overpowering the old pattern of movement with plenty of repetitions. This approach to motor learning basically went out of fashion in the 1970s, and has been replaced by an approach that credits the cognitive component of learning with much greater salience as opposed to rote repetition. That is, the athlete really needs to have an understanding of the motion pattern so he or she knows why the kinesthetic or biomechanical aspects of the motion need to be the way they are to produce the desired action. This approach treats elite skills as sort of channeled by the appropriate cognitive parameters of the movement. Because of this, the "old" pattern and the "new" pattern really aren't treated as necessarily in conflict. This shows itself in the "transfer" of skills, because skill for one distinct task does not generally transfer to another task. In a sense, the "new" motion is a "task" - do this in this way.
Another complication is the timing or tempo of the motion as a whole. Whether a motion pattern should be treated as a set of sub-motions or "parts" to practice separately depends somewhat on how the intended whole motion is structured in time and complexity. Complex motions that have clear sequencing of subparts can usually be practiced one part at a time, and later the parts are integrated. On the other hand, more unified, simple, quick motions may not be susceptible to breaking the motion down into separate components. Practicing "parts" in these cases can in fact be counterproductive to learning.
Richard Schmidt, in his Motor Learning and Performance (1991), uses the example of golf's full swing as a motion that might be harmed by learning the backswing separately. he says that the springlike effect of the top of the backswing, when performed as part of the whole motion, is different from the feeling when the backswing is practiced in isolation (pp 188-189). he considers the issue of subdividing the motion as one of whether the subdivision will lead to effective transfer of the practice to performance. He cites a couple of study in this connection:
Lersten, KC (1968). Transfer of movement components in a motor learning task. Research Quarterly, 39, 575-581.
Schmidt, RA & Young, DE (1987). Transfer of motor control in motor skill learning. In SM Cormier & JD Hagman (Eds.), Transfer of Learning (pp 47-79). Orlando, FL: Academic Press.
I don't fully agree with this, because neuroscience teaches the value of "end-point" positions of limbs in motor planning and limb movement. This may be a way to integrate parts into a whole, but in golf, there is always a going back and then a going forward, so there is more than one end-point.
Your example is the full swing. The putting stroke is arguably slower but less complex than the full swing. I don't know exactly how this cuts, but I have my current approach I can tell you about.
Generally, motor learning is said to occur in three stages: cognitive, associative, and automatic. There are different approaches to structuring practice as the student adfvances along this learning curve. Some include "segmentation," and "fractionation."
I believe that teaching a sound and effective putting stroke is usually a matter of ignoring the old pattern and starting fresh with the cognitive aspects of a sound straight stroke. By far, most golfers don't have very clear ideas of how the human body moves or what needs to happen in the putting stroke. So I explain the physics of what must happen for the ball to roll straight, and the biomechanics and kinesthetics that produce this straight roll in what I believe is an optimal way. There are a number of different aspects to this motion, and they can be subdivided more or less into the backstroke, the downstroke, and the finish stroke. There are ways to handle each of these phases or stages of the stroke, and I explain both the wrong ways and the correct or effective ways to handle these stages. This early stage of learning then is pretty heavily cognitive. There is a lot a talking and showing and letting the student try things out, and also letting the student just make mistakes and then explaining why the stroke produced the result it did and what needs to be done to avoid the problem.
During this stage, there is plenty of "parts" focus.
Once the student starts to get overloaded with this sort of cognitive stuff, I shift to the associative stage, where the focus is more on how the body moves in the whole sequence. Here, I emphasize tempo. There are specific cognitive things to work on for tempo as well, so this stage is really a mixed bag.
Getting the student to the automatic level of skill mastery, as I approach it, is giving the student a cognitive frame for the parts in the context of the whole, as well as strategies for integrating the parts into a fluid whole. The student is then on a self-teaching path with the ability to assess performance feedback accurately, to blame the correct aspects of the motion for poor results, and to focus practice on areas that really are weak.
So, in braod terms, I ignore the old pattern entirely. That said, I do know that all movement is influenced by stereotypical behaviors (opening doors, turning to the side, etc.) that we carry out day after day throughout our lives. Moreover, previously learned patterns of executing the skill tend to resurface and interfere with newer patterns in moments of inattention or lack of focus, as often arise under pressure. The traditional approach to this is to "extinguish" the old pattern. I personally don't really believe that rote repetition of the new motion is sufficient to extinguish the old pattern so that it won't resurface. Instead, I believe the athlete must have a belief structure that understands why the new pattern works and is superior to the old pattern, plus he or she needs to invoke the cognitive frame of the new motion practically every time. In other words, I'm not totally sold on the notion of automaticity, and instead belief that while attention resources diminish as the parts are learned and integrated into a whole, this does not mean that the cognitive aspects of the skills become transparent or less needed. I believe that inattention to these "procedural" aspects of the motion are a major source of error in performance, if not the largest source.
So, to get back to your question, I would concern myself with why the intermediate positions are functionally superior and how they relate to the whole in context. With this cognitive frame for the whole, you can pretty safely practice the parts in isolation, but don't allow this to convince you that you simply have to string everything together to get a satisfactory whole. In context of the whole motion, there may well be differences in the parts that only reveal themselves in the full action, and not sensing this would be counterproductive. Also, the only way to practice some aspects like tempo are in practicing the whole motion. Finally, once you feel the new pattern is sufficiently fluid, I would recommend continual revisiting of the cognitive or procedural rules for why the motion ought to be the way it is.
I hope this helps. I will try to beef up the motor skills learning resources in the science section.
Putting Theorist and Instructor
Golf's most advanced and comprehensive putting instruction.