The implementations in the benchmark are all pretty naive. You might get a different outcome with more careful implementations of the functions.