one way that comes to mind, it to use lwp to get each page, and benchmark the time it takes. not idea if the server is in use, as it lets other factors affect the times it takes to grab.
Main reason to grab them this way, is that it gives a slightly better idea of speed, taking into account how long it takes for the total request to go though, not just the time the program takes to run. (if an external binary is called to process the request, thats a factor that should be considered for speed purposes.) (apachebench may do this, I havn't looked at it)