...What? In MWI, interactions between "worlds" is the mechanism for all the quantum effects. If they never interacted, it'd give the wrong predictions.
Also, many proponents of MWI (e.g. Sean Carroll) don't think of the theory in terms of worlds at all. It's just a name that stuck. Really it's just what you get when you infer the Born rule from the unitarity postulate instead of having the Born rule as an extra postulate.
An intuition pump for this idea is: you probably think photons passing the point of no return and escape our Hubble bubble continue to exist. But you'll never ever see them again, even in principle! Wouldn't a model that says they simply get deleted out of reality be strictly simpler in terms of number of entities and therefore be preferred? Assuming a rule that simply deletes parts of the wave function that decohere away from us (i.e. collapse) is pretty analogous to that ( See: http://lesswrong.com/lw/pb/belief_in_the_implied_invisible/ ). Getting rid of the photon-deletion rule is analogous to accepting MWI (though I don't mean to imply it's the only game in town).
Framing the MWI as conceptually simpler than the other popular interpretations is a very interesting approach. I've read a few of Eliezer's essays along those lines, and I think he has a compelling point.
Now, you could argue that there's no test to distinguish it from other interpretations of quantum mechanics, but that applies to them as well.
> It's a mathematical model that makes observable predictions.
It sounds like you're talking about QM itself, not any interpretation.