"It's not so much that the CLR's limitations prevent it from running dynamic languages but that the CLR's limitations require you to invent a lot of your own infrastructure to run dynamic languages. If the CLR in itself assumes that it can resolve all method dispatches (or jump targets or attribute accesses) statically at compile time, you have to invent your own dynamicity atop that. If the CLR does not support first class functions, you have to invent your own approach. If the CLR does not support first-class continuations, you have to invent your own calling structure. Ditto named parameters, optional parameters, default parameters, and whatever other features that the CLR doesn't support.
I'm not saying that the CLR doesn't support all of those features -- I know that it does support some of them, to some degree. The DLR supports more. The question is whether Turing equivalence (and I hate this argument) is sufficient, or whether you're better off not inventing your own method dispatch system."