It's (supposedly) a quantification of the electrical leakage on the die.
In theory, an otherwise identical chip with less leakage should draw (slightly) less power to achieve the same clockspeed. In principle this should imply that the chip is more efficient, and may be a better clocker.
The results so far in this thread don't seem to support the theory though (mine certainly don't!). So, either the measure isn't very accurate, or other factors are more dominant - like, for example, the quality of thermal connection between the heatsink and the GPU.
It certainly doesn't seem to have any trend with higher numbers = better overclockers. I've tried it on a handful of 4xx cards now and all the good overclockers get upper 50s to low 70s and the average clockers are all in the upper 80s or upwards. Don't have any poor overclockers to hand to try - this is only about half a dozen cards tho so not enough to say that low number = better overclocker.