[voidlogic]

Well the best thing would be to try it yourself. I have found:

   * ab has more results variation between runs
   * ab will almost always report lower performance than wrk
   * If you have two implementations being benchmarked, A and B and B is always faster than A. wrk will report a greater degree of performance separation between A and B.

These results are less noticeable the lower performance the site being benchmarked is.

[justincormack]

ab takes more CPU time than a decent web server too.