It's actually quite funny... MultiThreadFailvoerTest was taking about 7 minutes to run before disabling JMX on the test, and 3 minutes after the change. (I ran all these without my changes for largeMessage & clustering) I have seen this MessageReferences following this path: MessagingReferenceImpl <- ConcurrentHashMaps <- QueueImpl <- MessageCounter <- ManagementImpl <- Somewhere_I_couldn't find (Probably JMX on the JVM) (This is a summarized graph of course.... the real profiler shows many more objects in between) I was only seeing MessageReferenceImpl growing on those tests, like the old JMXManagement Object was not being released. Maybe it would worth taking a look if management is not leaving objects after a server is stopped. View the original post : http://www.jboss.com/index.html?module=bb&op=viewtopic&p=4197452#... Reply to the post : http://www.jboss.com/index.html?module=bb&op=posting&mode=reply&a...