The current internal event bus - RepositoryChangeBus - starts a separate thread for each registered listener and offers events to each listener using a BlockingQueue. The listener in turn, polls the queue using BlockingQueue#poll, which seems to be quite CPU intensive under medium/heady load (http://supercoderz.in/2012/02/04/using-linkedblockingqueue-for-high-throughput-java-applications)

We should try to improve the performance in this area.