The only way anything can be gained by comparing scaling between a lower performing and a higher performing pair of cards is if CPU bottlenecking can be eliminated. Without enabling uber-high resolutions and uber-high settings on a massively overclocked system, all you are doing is gauging the bottlenecking level. The conclusion to that article confirms the point:
"In a nut shell, nVidia GeForce GTX 680 SLI scales really, really well. Overall perhaps a little bit less than AMD Radeon HD 7970 Crossfire, but in 5760x1080 it outperforms AMD's card. That's also the resolution where SLI truly comes into its own."
Tom's made that classic mistake in their comparison of scaling between the GTX 570 and 6950, two unevenly matched cards (similar to the GTX 680 vs. 7970). Of course, with less bottlenecking the 6950's "appeared" to scale better, and you know what happened... it was repeated ad nauseum as a plain and self-evident fact.
HardOCP learned this fact the hard way in their review of 6990 + 6970 Tri-Fire vs. GTX 580 Tri-SLI. Their initial review showed the 580's behind in scaling performance. Then, they did a second follow-up article in which they massively overclocked their Sandy Bridge CPU:
"In our original evaluation, every game we tested was a certain percentage faster on the AMD Radeon HD 6990+6970 Tri-Fire setup. With performance like that, we came to the conclusion that the $500 more expensive GTX 580 3-Way SLI setup was not a good value.
In our redux today, using our new 4.8GHz CPU, GTX 580 3-Way SLI has come out of its hibernation and is peeking around its fort ready to lay aim on the first prey it sees. That prey happens to be AMD Tri-Fire. In all of our testing today, the GTX 580 3-Way SLI setup performed faster than AMD Radeon HD 6990+6970 Tri-Fire, except for Dragon Age II."http://www.hardocp.com/article/2011/05/03/nvidia_3way_sli_amd_trifire_redux/6