Use one shared worker pool for all queues with 0-#CPU workers rather than separate pools per queue with 2-10*#CPU workers each.