> And even for background tasks shouldn't it be better to have them complete faster for less power

Race to idle is only clearly beneficial for tasks that have a clear start and end. If a background task is sustained, responds to unpredictable events, or does small amounts of work and wakes frequently, the CPU's boost logic won't solve your energy usage problem.

> To say nothing about what if they have different features. what happens when a process that wants to use cpu feature X(avx512?) gets scheduled on a cpu without X

This idea has been proposed in the past, but isn't actually used on x86-64 or ARM. E-cores have the same instruction set as P-cores, so there's no risk of running into an invalid CPU instruction.

Truly heterogeneous instruction sets may come back in the future, though. So be on your toes.