Very good article!
However, you could have benchmarked several more games in this comparison and one more mid-range GPU (say GTX 660M or ATI alternative) to emphasize the point. Hardly any gamer will look to spend enormous amounts of $$$ on a GTX 680M and complement it with intel's i5. Those that do get an i5 will chose a less expensive GPU such as GTX 650M, 660 or ATI alternative..
You failed to test the perhaps most important aspect here, physics heavy AAA multiplayer games such as Battlefield 3. The issue of inconsistency can be solved by joining servers running same map, with same amount of players (preferably full 64 people server), and taking averages over long games. A long conquest game can last from 40 minutes to an hour. Taking the average of the averages from 3 or 4 such games will ensure through the Law of Large Numbers that the result is consistent. Such information would be invaluable to see if it feasible to use dual core hyper threaded chips with a moderately capable graphics card.