Thanks to the discussion with @hft and thinking more carefully about the example given by Wikipedia, I think I understand now, but if I'm wrong then maybe an expert will correct me.
I think the point is that, after doing the Wick rotation, the equations and solutions that you get in Euclidean space do not have to exhibit the same features as the original ones in Minkowski space. If you interpret the Wick-rotated solution as describing a hypothetical physical world (a world with only space, no time), then that world need not even have the same qualitative phenomena as the world we perceive. That's why, in the Wikipedia example, a ball in free fall can turn into a hanging spring fixed at both ends, when Wick-rotated. Granted, certain generalities may have to be shared between the two, but in particular, and quite surprisingly, interference does not!
It doesn't matter, because the Wick-rotated Euclidean world is merely a staging ground in which the integral can be performed more easily; hence it need not look like our world. As A. Neumaier says in the original post I linked above, after you've done the integral you still have to Wick-rotate the solution back to Minkowski space to find out what phenomena we will actually perceive.
After all, even the most fundamental thing, the metric, is already dramatically different between the two spaces; a curve of constant interval is perceived as a circle in one, and a hyperbola in the other! So why shouldn't other features be dramatically different as well?
But I still can't help wondering, given that the Wick-rotated world seems so much more intuitive in the sense of having a simple Euclidean metric and simple additive probability, whether the Wick-rotated universe could somehow be the real one, and our perceptual mechanisms the true agents of rotation...