I mean, I don't really disagree; it's not a very scientific theory right now. It was just a blog post, after all. But if I was trying to test the theory, I would probably take a bunch of people who varied widely in writing skill and get them to write a short piece, and then get an external panel to grade the writing. Then I would get the same people to take some kind of test that judged ability to recognize rather than generate good writing (maybe get some panel of experts to provide some writing samples that were widely agreed to vary in writing quality, and have the participants rank them). Then I would see how much of the variation in writing skill was explained by the variation in ability to recognize good writing. If it was all or most of the variation, that would probably falsify the theory - the theory would say the most difficult part of "guess and check" is the guessing part, but those results would say it's the checking.

That's the first thing to come to mind, anyway.