[mgaunard]

The scheduling is based on how much the LWP made use of its previous time slices. A spinning program clearly is using every cycle it's given without yielding, and so you can clearly tell preemption should be minimized.