In my opinion, the benchmarks chosen look small (e.g. json parsing with 4096 chars, mergesort with 1024 elements, fib(20)).