Yea, but those kits are prohibitively expensive if I understand correctly.

Very few workloads are memory bandwidth bound. For workloads that are yes. You’ll get bigger gains by using all the memory channels and engaging all the memory controllers vs upping the clock.

yes, but throughput is rarely a bottleneck for consumer grade cpus. Maybe on the 16 core ryzens you could get some benefit for select workloads