A couple of people have mentioned on the users' list that 3.6 changes break CachedSqlEntityProcessor. Part of the problem is the current DIH tests are not realistic: they mostly test 1 class at a time and not the full end-to-end interactions between components.
This is a test of (Cached) SqlEntityProcessor using an in-memory Derby database. This hopefully is a first step in creating a series of new, more-rigorous DIH tests, which is needed in order to make changes without worry.
This also revealed a bug in SortedMapBackedCache, which is fixed also. I'm not sure this is the same bug causing problems for the 2 users but once committed I will post a reply on the users' list to see if they can try it.