Changing concurrency level didnt seem to help. However, If I get rid of the external transaction, it seems to solve the issue, even though there was no cache interaction in the external transaction at all, and it happens with some objects and works fine in some cases. I think when nested transactions are in play, there could be a problem in associating the right transaction to the lock, thereby causing this issue (the defect could be in jboss cache or atomikos or spring). I will see if I can put together a simpler program that can reproduce this problem and post it here. View the original post : http://www.jboss.org/index.html?module=bb&op=viewtopic&p=4250748#... Reply to the post : http://www.jboss.org/index.html?module=bb&op=posting&mode=reply&a...