beliefs may be epistemically irrational while being instrumentally rational.

I predict pushback -- LW won't like that idea one little bit :-)

and just accept the notion that there may be something interesting going on underneath the surface.

That's not the definition of "rational", though.

You're basically saying that with "irrational" beliefs one should go meta -- look at them from the outside, consider actual consequences and effects, ponder why these beliefs exists... Yeah, sure. Going "Oh, you believe in <deity>, you must be so dumb" is dumb. And yes, "I don't know why this works, but this works" is a reasonable :-) position to take.

I'm still not sure, though, that by the prediction metric science will look as badly as you hint and religions will shine. Contemporary religions generally shy away from testable (during the lifetime) predictions and for good reasons, too. As to science, I think the distinction between hard and soft sciences is pretty necessary here. Hard sciences live and die by predictions. Soft sciences spin narratives -- just like your friendly pastor next door :-/