May not get done for 4.
The Spring parsers for the various filters that rely on that <Entity>foo</Entity> style where it generates EntityIdPredicate objects around those elements aren't optimal. Instead of building one Predicate around a number of Entity elements so the eventual test is one check against a Set<String>, it turns into N Predicates with a single String, or'd together. So the check is O(N) instead of a more optimal HashSet lookup.
There are some use cases where there's no practical way to express the proper set of SPs except a big hellacious list, so this could be meaningful in those cases. Algorithms come to mind.