All Load Balancing Strategies should be CPU frequency (rate) aware
Under an assumption of full or near-full CPU load when applications and the RTS are running well, we've found that TurboBoost and similar dynamic overclocking systems produce a wide range of resulting speeds across the many processors in a large cluster/supercomputer. Thus, each hardware unit will be capable of different amounts of work over the same unit of time. Our load balancing strategies should uniformly take this into account.
There's a higher-order bit of adaptivity that's possible here, of strategies intentionally un-loading some cores to allow others to boost further. I'm not asking for that here.