I think the whole point is that microbenchmarks provide no data in overall performance. They just test very specific and by no means common use case.