Have you tried with larger input data. Whilst many itterations is a good quick test I think what you are dealing with is a n^2 problem (maybe higher) and so a larger initial data set may spread the results more and give a clearer identification of the best algorithm