I find it more reasonable to assume that you can only improve a very specific function so much before your "improvements" turn to "pointless sidegrades".