Assume a scale from 1 to 10. That gives us a matrix of 1,000 potential points in the GNS spectrum. We don't have to design 1,000 games, but we could probably design 100 games that would be statistically representative of the whole matrix. "Games" in this case meaning "enough game to be played and graded by test groups"; not necessarily a complete game designed to be played over a number of sessions, or even pushed really hard mechanically.
Now, induce a large number of independent game groups to test these games, and rate them. Before they "qualify" to partake in the test, you do another test on those individuals to segment them psychographically.
Now you compare ratings on the games to the psychographic profiles of the players. You're looking for surprises - combinations of G, N & S that trigger high ratings, or low ratings. Finding one or two of the segments like a particular combination will not be rare. Finding a place where all 4 + the Basic Gamers respond strongly will be rare. But once found, you would have the blueprint for a game system that would likely be very competitive in the market.
Likewise, if you find things where everyone expresses dislike, you can learn a lot from that too. Studying the failures may teach you a lot about things to be sure you avoid in development -- and you may find things that are not conventional wisdom; i.e. new knowledge that we haven't had before which can be fed back into the R&D cycle to generate overall improvements, even in existing games.
That's the kind of research I'd do if I had a million dollar pure RPG R&D budget at my disposal.
Ryan