So I've been thinking about this recently for some reason (I guess I'm always thinking about issues surrounding realism and anti-realism). In his discussion of Putnam's paradox, David Lewis wonders briefly why Putnam supposes that only an ideal theory is guaranteed reference and truth, when a version of the model theoretic argument can show that any theory is guaranteed reference and truth. He supposes that Putnam is assuming that our theories are forward-looking, so that their intended interpretations are as according to some future ideal theory.
This seems obviously right to me. However, I think it shouldn't have been that hard to get to that point. The lesson Lewis ultimately draws from Putnam's paradox is that global descriptivism can't work; the meanings of the entirety of our language and theories can't be determined by picking what would make them as true as possible, because a cunning interpretation can make virtually any theory true of virtually anything. This is despite Lewis's enthusiasm for local descriptivism; for individual words and concepts, it is a good approach to interpret them to refer to whatever makes the most, or at least the most important, of our beliefs about them true.
That strikes me as absolutely the point of Putnam's paradox. We can always interpret some part of our theory in light of other parts of our theory. This is surely why Putnam only draws the conclusion that an ideal theory must be true. Any less than ideal theory can always be interpreted in terms of future, more extensive theories, and can turn out to be partly wrong on the basis of that latter interpretation; only a theory that is as inclusive as possible is immune to being so interpreted by further theory.
Of course, from the conclusion that global descriptivism is false, Lewis moves on to maintain that we need some constraints on interpretation beyond making our theory, whatever it is, come out as true as possible. Infamously, he maintains that theories should be interpreted as much as possible as referring to his perfectly natural properties and things which are built out of them. The Canberra Credo maintains that Lewis was right about pretty much everything, except modal realism. If I were to construct a credo, it would say that Lewis was right about pretty much everything, except perfectly natural properties.
Instead, I would say that it makes no sense to try to interpret a complete, ideal theory; all interpretation is internal to our theorizing, and whenever we're evaluating some theory, it is always in light of some further theoretical commitments. I take it that this was the real point of Putnam's internal realism, as well as the old Logical Positivist rejection of metaphysics (I assume Putnam's "internal realism" terminology came from Carnap's distinction between internal and external questions; Carnap of course rejected external questions).
I should probably look up van Fraassen's paper on Putnam's paradox as well. I'm usually in sympathy with van Fraassen.