Well these processors have pipelines of say 15-30 stages or whatever. That means that an instruction has a 30 cycle turnaround time. Of course, the instructions are 'pipelined' so that many are put into the pipe one after the other. There is branch prediction logic to know what future instructions to put into the pipe.
Of course mistakes will be made and then there is a lag to bypass those erroneous instructions that were pipelined. So the performance can differ depending on the software itself. I think that usually the clock speed and the pipeline length are good indicators of the processor speed.
Notably, the earlier Athlons had a much shorter pipeline than the P4's which led to them having a PR rating, like 1700+ which actually ran at 1200Mhz or whatever, don't quote that figure.