Non shared stores should allow for segment based store separation. This might involve creating a store per segment. But this allows for superior iteration performance over a subset of segments.
Distributed stores benefit the most from this due to the fact that iteration is done at the segment level to help ensure that duplicate entries are not retrieved. It also would be beneficial for state transfer and other operations that operate only upon a given set of segments.
This could be advantageous even for REPL and LOCAL caches as there is then a very clear separation of stores to process in parallel for given operations. This would have to be verified with tests to see if this is worth it though.