There is a subtle race condition in JMSClientInterceptor: ResultJoinHandler handler = createResultJoinHandler(methodInvocation, metadata); // send request to topic requestor.request(destination, requestMessage, handler, getMultipleResponseTimeout()); // race to here if the response comes back fast RemoteInvocationResult result = handler.waitForResult(); return recreateRemoteInvocationResult(result); due to a bug in the ResultJoinHandler methods: public RemoteInvocationResult waitForResult() { while (true) { synchronized (lock) { if (result != null) { return result; } If the onMessage result could have been already called before we call waitForResult and so result could be != null but the unblockCallerThread could still return false. I've attached a patch which fixes the problem. Few of the unit tests work for me because of outside dependencies so I'm not sure how to best test this issue.