On the other end of the spectrum, systems with “bursty” load for which traffic patterns fluctuate drastically may do better with a queue size based on the current number of threads in use, processing time for each request, and the size and frequency of bursts.
