Is the thrust of the thought experiment preserved if we assume that the two versions of water differ on a chemical level, but magically act identically on the macro scale, and in fact are identical except to certain tests that are, conveniently, beyond the technological knowledge of the time period? (Assuming we are allowed to set the thought experiment in the past.)

Surely it's not necessary that the two worlds be completely indistinguishable?

It doesn't behave just like water. It behaves like a simpler model of water. If you look more closely, the difference isn't what you'd expect between a good model of water and a bad model of water. It's what you'd expect between a good model of XeYZn and a bad model of XeYZn.

In other words, it would act like water to a first approximation, but instead of adding the terms you'd expect to make it more accurate, you add the terms you'd use to make an approximation of XeYZn more accurate.