47/59 vs 57/70, just for 2 extra cores and an extra slab of cache. That is a pretty clear indicator that moar cores is a good idea, alpha or no.
I just dont see it quite so clearly that way. For one the clockspeeds are different, but if you compare 2500k to 3930k, which is closest in clockspeed, the difference in ave fps is only about 7.5 percent and about 5 percent in min fps. And if you compare HT 2600k tp 3930k, the difference is probably within the margin of error. Granted 3970k is 10 percent faster, but it has more cache and slightly faster clocks as well.
If you go to the article itself and look at the cpu thread usage for the 3970x, only 4 real cores are loaded above 40 percent, along with one virtual core. Actually, the core usage looks strange, in that one real core is not loaded at all and another is only 23%, while a virtual core is loaded nearly 50 percent.
BTW, as an aside, this site always mystifies me, in that they seem to have so many games to test but are still using SB for the intel processors. Not that it would make much difference, but for a scientific comparison one should compare the latest processors from each manufacturer. And actually over the two generations from SB to Haswell, you might see a moderately improved performance.