Currently, the multiple-producer-single-consumer queue used in TransferQueueBundler is ArrayBlockingQueue. Try the following to see if it improves performance (BundlerStressTest?):

• LinkedTransferQueue: unbounded, but put() could be overriden in a subclass to either discard the element (similar to RED), or block, when the queue is full
• LinkedBlockingQueue
• ConcurrentLinkedQueue(2): remove the queue impl which isn't used by TransferQueueBundler