CPU/GPU performance is limited by lots of things. Clock speed is easy to measure, but isn't the only factor to affect performance. Pipeline length (e.g. how many steps are required from start to finish for an instruction to complete), multi-threading, branch prediction, etc, all contribute to real world performance.
Ars Technica had a decent intro on this, but obviously it only scratches the surface.
As for what spec to check first, if you're looking for the best performance for your needs, I'd look for benchmarks, not just manufacturer specs. Which benchmarks are of value to you may be very different depending on your planned application, so you'll need to exercise some discretion.