You've got the gist of it. They use a Gaussian distribution to weight the likely contribution of each recruit at that rating level compared to the other members of the class, without consideration of the rest of the roster (class=team). They assume the ratings are accurate (obviously) and then figure (proprietary) what kind of contribution each player than make: your best player will have the biggest impact and will be the most involved. So if you have two 4* guys, the model may say they both play 50% of the time. But the team with a 5* and a 3* might be getting 80% of playing time from the 5* and only 20% from the 3*. Therefore, it doesn't matter that they average to a 4*, the 3* won't be around as much.
You can extrapolate the recruiting classes to the 85 team limit. If UGA gets six 5* per class, they will average twenty-four 5* (ignore 5th years and early option on draft eligibility since we're talking about recruiting a team for it's ability to play) on a roster. If OSU only gets three 5*, they will only have 12 on the roster. Therefore, UGA will be able to field an entire team of 5*s and OSU will only be able to field half a team. Therefore, it doesn't matter than UGA also has thirty 3*s that bring down the average...they barely factor into the equation.
Of course, the model needs to consider the value of position depth (having a middle of a roster that's just as good as the top of the roster) and how all that talent stacks up by position (if all the 5* are QBs, it doesn't matter)...but that's a decision 247sports needs to make based on what they think is important.