Turning up userCount under REPEATABLE_READ and PESSIMISTIC locking yielded the same failure. So, looks like there are 2 distinct problems: 1) One we've been debugging all along: cache left in incorrect state when one thread wants to remove node while another wants to putForExternalRead (not clear whether the problem is that latter wins or former fails on invocation of RemoveNodeCommand) 2) Under REPEATABLE_READ + MVCC: CacheInvocationDelegate.removeNode(Fqn) returns false and fails to remove() node in question. This happens w/o any concurrency (single thread) Thought latter seems easier to debug, I can (prefer to) live w/o REPEATABLE_READ, so still problem #1 is the one to solve for me ;- View the original post : http://www.jboss.org/index.html?module=bb&op=viewtopic&p=4215879#... Reply to the post : http://www.jboss.org/index.html?module=bb&op=posting&mode=reply&a...