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.