This comes hot on the heels of Thread Pools on the JVM[1], which suggested breaking up thread pools into cpu bound, i/o bound, and latency critical. To that effect, not sure what the guidance is here: what are good practices in Rust when you have different workloads running alongside each other?
Write-up looks good though, some good suggestions.
[1] https://news.ycombinator.com/item?id=27883847 , ~115 comments