Each workspace cache is currently a simple concurrent map, and because it is shared by multiple sessions it is never cleared nor its contents removed. This is obviously a problem that needs to be addressed.

Perhaps the easiest way to do this is to use an in-memory Infinispan cache for each workspace cache, where the Infinispan cache has no cache store, limited numbers of entries, with a LRU or LIRS eviction strategy and possibly even a lifespan limit. (Not sure how/whether to expose these settings in the repository configuration; probably a good idea to better support deployment on various sized machines and custom tuning.)