Needless to say, all speculation:
It's pretty hard to be dumb as a post and score 1500+ (excluding the short-lived essay). It may not be the most predictive factor for college success overall, but as the top schools have become increasingly competitive, they can afford to use higher and higher scores as a baseline. If they miss some diamonds in the rough who test poorly, they still won't have any trouble filling a class. Test-taking wizards with little else to recommend them will get filtered out through other methods.
On the other hand, if you're a mid-tier school trying to choose between a pair of otherwise similar applicants with scores of 1000 and 1200, it might matter to you a great deal how predictive the SAT is vs. the ACT or some other metric. Meanwhile, no matter how confident you are in its predictive power, at least it's an objective metric. One that, for better or worse, also has a marketing impact, since average incoming class scores show up near the top of college survey data.
I think most of the controversy about the SAT and other standardized tests centers around the predictive power of results near the median. Especially since re-centering, the test doesn't try hard to distinguish the top 1% from the top .01%, but it's very important that it correctly distinguish the 60th percentile from the 40th.