that code was indeed quite complicated (it seemed to wander back and
forth between draw, mouse and screen in a way that was probably straightforward
but always left me a little confused for a few passes).

for performance, i think it depends how it's done.  i didn't do the Inferno one, but i did
use it without obvious trouble with Inferno on a 486sx33 processor.  in one of my
own os's i did the software cursor in a process on a 386sx16 and
that was fine too, although that was only 16x16 in bits on a monochrome screen.
neither controlled it with the clock, though, which might have made it smoother.