The power-aware scheduling mini-summit

I'm not seeing anything about task grouping (i.e., grouping timer-based wakeups so as to cause as few movements from idle as possible, within reason). That seems a pretty obvious (even if not obvious how to actually accomplish) place to look to go into deeper C-states.
Perhaps this has been mentioned elsewhere?
Also, has anyone considered creating a few "simple" workloads that run on an ideal scheduler? That is, the scheduler knows "exactly" (to slice precision) how, and how long, each task will run. That seems a reasonable metric to gauge further progress on the scheduler.