"Various studies conducted over the lifetime of the SAT show a statistically significant increase in correlation of high school grades and freshman grades when the SAT is factored in." from wikipedia http://en.wikipedia.org/wiki/SAT
They mention that a 100 point increase on the SAT corresponds to a 0.1 boost on college GPA. Assume a linear relation: this means that a 1500 SAT student might expect a 3.0 GPA, while a 1000 SAT student might expect a 2.5 GPA. That's actually a great predictor.
Then they mention other studies of SAT scores within a fixed college. This is deceptive. A typical college has students from a narrow SAT range (say 1200-1400). By the above numbers, one expects that SAT predicts only 0.1 GPA above/below the mean (within the college). However, it also predicts that if the school were to ignore SAT and admit a cohort with SAT's of 800, then those students would underperform by half a letter grade.
If you really wanted to figure out the value of the SAT you would have to use a much more robust measure of cognitive development than freshman GPA. However, it doesn't look like anyone is going to undertake such a massive project anytime soon. Incidentally, Alfie Kohn has some really good stuff to say about grades in both Punished By Rewards and also in What Does It Mean To Be Well Educated?
Also, I should have mentioned above that there are some meta-studies that find that degree of validity depends on the person's race, and that if you take the person's race into account the test can be useful. I have multiple objections to this though, both practical and theoretical.