[fizx]

Large queues can have problems. Perhaps your queue is effectively 30 seconds deep at times. But your clients time out around 10 seconds. Now every request in the queue is useless. If your clients retry, then now you have a feedback loop generating uselessness. You'd be better off with a short queue, and rejecting requests much faster.

[IgorPartola]

Browsers don't automatically retry AFAIK, and users are not likely to hit refresh after seeing a hard error like connection refused. If you reach a queue that gets to be 30 seconds long, you most likely need more/more powerful backends.