So, not a teacher, but I have a lot of close friends who are teachers; this post is mostly informed by what they've told me about their experiences. First, I would say that segregation by ability is common in high school, but uncommon in elementary school. It can go either way in middle school.
So, in my mind, your question is mainly asking if increased amounts of differentiated instruction could benefit elementary school students. I've definitely heard way more about the difficulty of dealing with poorly behaved students than the difficulty of dealing with well-behaved, but low achieving students. That suggests segregating not along ability lines but rather along behavior/self-control lines.
If you run with that idea, and keep class sizes the same for well-behaved children and poorly behaved children, the "good" kids would probably benefit, but the "bad" classrooms would be completely out of control; in effect you would be completely writing those kids off at a very early age. Depending on the capability of the teachers and the difficulty of the student population, I personally think that might actually be a socially optimal use of resources. However, that is a very controversial opinion, and there are a whole lot of people who would not agree with me on that (and most of my teacher friends are among them).
If you go to very large classes for "good" kids, and very small classes for "bad" kids (say ~50 kids vs ~10), you have a different set of problems. A class of 50 8 year olds can get pretty hard to control, even if they're well-behaved 8 year olds. Plus, most schools don't have the facilities to accommodate classes that size. You can bet parents aren't going to be that pleased to have their kids in huge classes. And even if more individual attention is what the "bad" kids need, you still have the problem that you've labeled the bad kids as bad; they'll figure it out and it will probably have adverse affects on those kids' development. That may offset any benefit you're getting from the smaller class size in the "bad" population.
So I don't think it's actually the slam dunk it might seem from a casual analysis.