On an individual basis. But who really cares about that? There's 85 guys on a football team, and if you know all the probabilities based on the stars, you can probably predict how many all americans the team will have pretty well.
I am not exactly sure what your point is. But looking at the graphic in kchusker_chris’s post above, if we recruit a 5* guy he has a 1/13 chance of becoming an All American. A 4* guy has a 1/53 chance. And a 3* guy has a 1/172 chance. Sure, you could use those odds to figure out how many guys on the team should eventually become All Americans. Since we don’t care about individuals or anything.
But the OP’s post is correct also: two-thirds of the first-team All Americans were not among Rivals.com's Top 100 when they were recruited. This is mostly because the pool of 3* players (4,982) is so much larger than the pool of 4* players (1805) or 5* players (171). But it's also because evaluating talent, projecting a guy's improvement, and predicting whether he will shoot himself in the foot and derail his career are difficult things to do. In other words, predicting the future is a difficult thing to do.
My point was just that predicting the future can be an easy thing to do. It just depends what you're trying to predict. Example:
"Dorial Green-Beckham will be an All-American one day" = Hard to predict
"One of the top 13 will be an All-American one day" = Easy to predict
Which means evaluating the ability of a team in the future is easy to predict. When evaluating a team, you don't really care who the contributors are...you just care how many there are and how good they are. Look at 85 guys, sort them by star ranking, and you'll have a pretty good idea of what their overall output will be in the future (i.e., wins). What you can't predict is which specific player will be the best, but that's not the question we're really interested in answering anyway.