I'm not convinced that when you look at the whole set of minds doing dovetailing simulations and put probability distributions on how far they go, your algorithm and Eliezer's give different results. Actually calculating it out looks a bit tough; my intuition is based on the fact that a simulator doing N computations gives program n of the order of Sqrt(N) computations using either algorithm, provided that n << N.
Well, since any finite number is smaller than infinity, for ANY program, once it starts running, it would get just as many steps per, well, step, as any other program (in the original version). ie, consider two programs A and B such that A came earlier. In the original scheduler, once B started up, for each tick A gets, B would also get one tick. But A would also have an initial bunch of ticks that it got before B even started.
My version makes sure that B gets those extra ticks too, that's all. I, personally, don't think it would change the probability distributions that would be experienced from the inside, given that the base computation really is run with unbounded resources and so on and so forth.
So I'd intended this story as a bit of utterly deranged fun, but it got out of control and ended up as a deep philosophical exploration, and now those of you who care will have to wade through the insanity. I'm sorry. I just can't seem to help myself.
I know that writing crossover fanfiction is considered one of the lower levels to which an author can sink. Alas, I've always been a sucker for audacity, and I am the sort of person who couldn't resist trying to top the entire... but never mind, you can see for yourself.
Click on to read my latest story and first fanfiction, a Vernor Vinge x Greg Egan crackfic.