Precisely. I was going to further add that most load balancers are network appliances and hence: a) There must be network packet buffers on each receiver, whic

Precisely. I was going to further add that most load balancers are network appliances and hence: a) There must be network packet buffers on each receiver, which is a per-server "queue". b) The optimal performance can only be achieved by pipelining requests sent to the servers, otherwise there is "dead time" caused by the network latency. Perfect queuing to workers that act as a "slot" can only be implemented in-process on a single machine, and even there having small queues or buffers per core can improve throughput!