ab is not a right tool to measure anything close to reality. It runs the same HTTP request multiple times which effectively re-uses caches (file system's and db's) starting with the 2nd request, so the ab results are always screwed.
Regarding xdebug, if you can do that in this tool, why do you need the same functionality in dbg? I think we better focus on something unique that is not available in xdebug.