Would a better benchmark not just use some kind of pseudo randomly generated sequence to avoid having two arrays?