Does your model differ from "Everything that could exist, does"?

Not quite. Every mathematical structure (strictly, every syntactic language) that could exist, does (and causes its corresponding apparent physical reality to appear to exist) - so the kangaroos do indeed exist (and did so even before you thought of them). After all, how could you have an effect on the kangaroos merely by simulating them? (I maintain that the causal relation runs from the kangaroos to you, not the other way round).

It may well be impossible to devise an example of something that doesn't exist (because, if we can think about it, it is modellable), but that doesn't mean to say that things that don't exist, um, don't exist... I mean, {X : not exist(X)} need not be empty, it's just that we can't exhibit any of its elements. It might be provably empty, but I have no proof here and can't imagine what form one would take.

Also, you have to be a bit careful about the causal relations; if you start thinking of them as one formal system modelling another, you get into Gödel-space. Hence the Syntacticism post; I think it's accurate to say that one formal system models a quotation of another, but I'm not sure of that.