It doesn't work like this though. Even if you take the 50% performance increase on face value (not taking into account higher AA/ASF/Shaders) that would mean a game running at 15fps would increase to 23fps. Not exactly much of an increase. Even if you were getting 30fps on the GTX 260, that's an increase of 15fps (which is what the tests essentially saw), hardly worth $300.
Meanwhile, if you spent the money on CPU/MBD/RAM & a mid range graphics card (say a GTX 480 at around $150), you'd see actual performance increases of around 3.5x that of sticking a GTX 660 on a crap motherboard with a crap processor.
Sure, if you had every intention of upgrading the rest of the components, the graphics card is going to be the easiest to swap out, but you're still going to need to upgrade the CPU/MBD/RAM.
The article hides the fact that the increase of a GTX 260 vs GTX 660 card in a modern system would be a ~400% increase in performance. Not sure what they're trying to prove, but to me it proves they know nothing about hardware, gaming or value for money.