[a1369209993]

Except that's not opposition? Microbenchmarks do, always, lie; you need to measure the performance of your actual, real code that you're actually caring about the speed of.