Cut the number of split()s in 1/2, Almost cuts run time by half.

Instead of splitting large data sets once for every percentile
calculation, do it once, then use the result split for all the
percentiles.