Timeslice scaling is probably a mistake
The smart scheduler algorithm dates to around 2000, and computers have changed a bit since then. The kernel will now happily deliver timer signals as finely as 1ms, and the vast majority of X requests will take much less time than that. Stretching a single client's timeslice when no other clients are active now mostly means we will be slower to respond when the next client becomes active, because we won't decide to switch which client we're servicing until after the timeslice expires.