On multicore machines you want to process timers in parallel on multiple cores.

By experience I almost never want to run my timer callbacks on a random core and always want it on a specific core. With the typical one event loop per thread, you would register the timer with the thread you care about.

This is going to be very application specific.