OC numbers don't really make sense when you consider all the variables involved. You might find that a nice hsf will outperform many a good wb simply because you're cooling the fets more. We really don't even know all the reasons why a chip will clock higher under certain conditions than under others. For the ones that we know of, there's already cpu temp, voltage, voltage stability, fet temp... even some weird phenomenon that could probably be ignored (but who knows?) such as electromigration, alpha particles, and cosmic rays. AND the relation between the factors is not simple.
I'm not saying it can't be done, but you must construct a system where at the very least, OCs vary predictably with CPU temps.
jaydee, not sure exactly what you mean by "react." Results from a single CPU are no more indicative of the performance on other CPU designs than are results from a die sim. Maybe worse. Even for dies with similar geometry, every design has different hotspots.
Here's a suggestion: poll your readers (not just the forum nerds) to see what they are doing. What CPU, willing to remove IHS, current cooling, etc? But keep it short and easy! You want to get as large a sample as possible.
I guess a prerequisite to this is to *have* readers or even articles to read for that matter 
. So once you get an article you're certain will receive fairly wide circulation, *make* them take the survey before the last page or something along those lines. Joe's smart, I'm sure he can figure something clever out. And I hear he doesn't mind extra coding at all! Then you'll know exactly who you're readers are, and you can go from there trying to determine how best to serve them. JoeC recognized this fundamental issue long ago on oc.com.