The mean IQ of college grads started out as higher than the mean IQ of the overall population. As you add more people from the overall population, you would expect the college grad mean IQ to get closer to the overall population mean IQ.

Right, if your situation is just "X is the IQ cutoff for getting into college," you could infer that as more people go to college, X gets lower. But you could also see a situation in which smart people opt out of college, so even if the number of people who wanted to go to school were constant, the average IQ would drop -- e.g. fifty years ago a smart person could succeed by getting a scholarship to a great school; now said person gets 90% of what he needs to know from Wikipedia, and can spread a great idea by uploading something to Sourceforge rather than muddling through the thesis-writing, thesis-approving process.

I don't think it's just one phenomenon, and a model that either considers America a pure IQ meritocracy or assumes that college's role has changed by degree and not kind is clearly inadequate.