I worked with AIX on Power5/5+ (mid 2000s) and their hypervisor for LPARs and what they called micro-LPARs at the time was astonishingly good. It provided very nice granularity and control for running multiple LPARs. An LPAR is like a VM, but is closer to bare-metal than what you find in an Intel environment. You had to allocate at least 1/10th of a virtual processor per LPAR. They had something called 'entitled capacity' where you could specify the minimum amount of CPU allocation the LPAR would get even if all LPARs running on the same hardware were running 100% busy at the same time.

It all worked exactly as advertised. I was on a small team that did performance monitoring, optimizations, and capacity planning and we watched the metrics very closely for production workloads (running DB2). It was seriously impressive.