With the addition of the event bus, each listener registered with the bus is notified from a separate thread, for each workspace. If a separate thread(s) is used for the system workspace as well, at some point the node caches become invalid (randomly).

To prevent this issue, a temporary boolean configuration flag has been added to the RepositoryChangeBus - separateThreadForSystemWorkspace - which is set to "false".

To reproduce this issue, change the above mentioned flag to "true" and run several time the JcrVersioningTest suite. Random failures should start to creep up.