I'm going to release a big update very soon, but first i'd like to comment something on the quad mode algo.
At last i think i know how the original double one worked. It copied the first half of the screen onto the second, i got it right after all (copying half of the pixels) but i could also have copied the first quarter onto the third and the second onto fourth, performance wise it could have been even worse. It's very difficult to see what was going on there without any commentary.
On a pentium mmx rendering interpolation it's extremely slow (1 fps), it wasn't on 2.2 and i don't know where to look at.
AFAIK Graf Zahl did a full interpolation rewrite but doesn't seem mmx is used anywere, on a pentium or pentium 2 it's fine.
Direct3D will be working on the next release BTW.