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