The following is a sample idea
A scheduler that knows which processes are related to which server port and can therefore directly schedule (or prioritise) processes that handle a specific networking packet. This should reduce latency.
This could be it. But don't feel shy to add a bit more explanation.
A task scheduler for the parallel workloads running inside a guest that looks up the value of N = the vm's cpu.max.concurrency [1] field provided by the host at every timeslice expiration, and places tasks on 0-N vcpus, in order to match the number of active vcpus on the host with the value of N.