Thanks liquidfire, for your simulation. Good idea. It's the next best thing to me actually making the change in my pool, to test it out.

There is a problem with it though. Block solve times, and share solve times, are not between 0.5 and 1.5 of their average. Nor are they uniformly distributed. They are between 0 and infinity, and some kind of inverse square distribution maybe.

I believe that bug is what's producing your results, showing that slow miners are disadvantaged. Because if a share cannot be solved in 0.5, and a block comes before then, then his work is wasted.