Perhaps a headcount-based approach is flawed, to the extent that they shouldn’t build infrastructure recklessly and enroll beyond their capacity to graduate students. Burlington College and its recent closure, for example, may be a cautionary tale of overextension with regards to infrastructure. The space available is not constant in the long run, and is a factor of how much infrastructure the school chooses to build. And while there is some potential to retool infrastructure for other purposes (housing and academic buildings have other interested customers for their services other than college students), it should not be built irresponsibly.
The predictive power of high school factors is much stronger on the lower end than it is on the upper end. That is, it may be pretty hard to predict who the best students are, but it’s not so hard to predict who is a bad risk. A high HS GPA and a high SAT do not guarantee good results, but a low HS GPA and SAT are very accurate guarantees of a likely bad result. The latter group should not be admitted to an expensive university program because if there were any good degree of financial liability on the part of the university for bad enrollments, that would be written off as a bad risk. And for the fringe cases where that proves to be an inaccurate judgment, there should be a backdoor admission through something like community college.