This was based on the assumption that queue would be trimmed fast, leading to a tight loop. We now go one round on existing queues.