Here's the correct answer: Kepler's load power draw isn't static - it is application dependent, depending on the final clockspeed and the power % used during the 3d application. The load power draw isn't static with the Kepler architecture, which is why many websites are aggregating samples with various 3d game power consumption. Overclocks also heavily weigh on the power output.
HardOCP is a great example. Look at how they measure power draw - actual 3d applications, taken from a sample of 3-4 games. Why? Because power draw isn't static with Kepler. That said, I really give no cares whether my 780 draws 250W or 300W, and I imagine that it won't be a big deal with the 9970 either. Note that the higher GK110 is overclocked, that makes the power consumption go way up as well. But, who cares? Really? The only thing *I* care about are the secondary effect - noise. Most of the aftermarket 780s are really good in this respect, and AMD is rumored to be working on an improved reference cooler with the 9970. The 7970 ref cooler really wasn't as good as it should have been in that respect (eg noise). I suspect that once you get into the range of 500$ GPUs, most people are running hefty PSUs and care more about secondary effects such as heat and noise rather than what their electric meter says.