When requests arrive faster than they can be served, where does the excess end up? Load at 125% of capacity, with nothing to absorb it. An unbounded queue turns the excess into latency, a bounded one into fast errors, and blocking hides it in the callers. The demo needs JavaScript.
Past 100% busy the excess becomes queue, errors, or blocked callers. A queue policy decides which of those you get, not whether you get one.
A queue limit is a wait limit: the limit divided by throughput. Keep it below the client timeout, or the queue fills with work nobody is waiting for and the limit only rejects what is already lost.
Backpressure helps only if the caller can slow its own source. Here it cannot, so the backlog sits in threads and connections upstream, where the server's dashboards cannot see it.
All demos