Let's play devil's advocate: you have limited and messy data about the qualifications of a hundred 20 year old candidates. You have one qualification that shows they worked quite hard for the majority of their life and a sustained and high level. You should value that data point almost as much as you value hiring the best candidate. There is not likely to be any other data point as valuable at that point in the applicants' lives.
It's a Bayesian problem all the way through and you can probably quantify the value of the signal provided. However, and most importantly, it's just a signal. I personally prefer hiring interns, completely irregardless of school, and developing and hiring them, any time. At that point, there isn't signal, it's just measured performance that is de-risked.
Now, people that use it as credentialing are morons. I mean that bluntly and literally. Randomly picking on CFOs: "I want a CFO with a Harvard degree" is stupid thing to say, on its face. The person saying something like that has literally inverted the real point, which should be, "I want an excellent CFO, and a Harvard degree is a signal of some properties correlated with that outcome". Once you invert that, it's just another data point in a meritocratic decision, and a relatively minor one.
Sorry to post/write so much on this. Startups are the ultimate meritocracy in some ways, and having something posted here that's populist and almost anti-intellectual bothers me. Ycombinator has no problem pulling from Stanford. And Stanford, MIT, and Duke (and many others) are as elitist as any Ivy on any day that ends in y. That's leaving out the really class conscious "small liberal arts" colleges out there, and ignores the differences between ivy league schools in this regard, which is large.