Hey, not to be a bad apple, but statistically speaking, assuming that your data (time for runs) follows a semi-normal curve, you would need a minimum sample size of 25/set...but that's just way too much of a pain in the butt. If you consider the problem as a probability set, where the chance of one set being better than the other, then n*p and n*q have to be greater than 10, where n = number of runs, p=success rate of a particular set, and q= fail rate of a particular set. Otherwise, data is statistically inconclusive.....according to my AP Stat class...Good job though!