Right, but your arguments against OCing apply to OCing across various systems.
Let's take a *fixed* system, say an AMD one, with *fixed* mobo (say an NForce2 for max potential) and *fixed* CPU. Of course that reference CPU will have to be one that gains extra MHz when the temp drops by 1°C. Do not change anything but the waterblock.
If you cannot get consistent temps, what's the point in measuring temps or C/W anyway ?
So let's admit you manage to get consistent temps (from controlling ambient for example, measuring it precisely is easy). The extra measurement is CPU temp, but on that 'reference' CPU it's correlated to max overclock. What is the easier measurement ? To me it's max overclock. Even if the relation between the two is not linear, who cares.
Remember, that is when *all* other parameters are fixed, otherwise nothing is possible.. but that holds for any benchmark / temp measurement / etc.
Oh and keep a small temp somewhere, just to detect immediately those "piece-of-crap" blocks... if the mobo sensor displays 80°C there's no point going any further
(edit) ah and the variations between OCing runs (with the same system) will be due to mounting and TIM. No need to add temp probe randomness to that...