1. Let's not bring Exalytics into the equation yet -- at this point I'm going to withhold comments for various reasons...
2. I've tested large ASO models and have seen no benefit in allocating more cache than is recommended by the DBA Guide. I'd love to see the cube that proves otherwise. The bottom line is that cache in ASO isn't very important.
"A 32 MB cache setting supports a database with approximately 2 GB of input-level data. If the
input-level data size is greater than 2 GB by some factor, the aggregate storage cache can be
increased by the square root of the factor. For example, if the input-level data size is 3 GB
(2 GB * 1.5), multiply the aggregate storage cache size of 32 MB by the square root of 1.5, and
set the aggregate cache size to the result: 39.04 MB."