Well the replacement of time by imaginary time makes the transition from QFT to statistical mechanics and the other way around. I mean, we know it works but we do not know if there might be some underlying system that governs this behaviour...I mean, this is a bit like working with complex numbers : in order to describe something real in an easy way, it is favourable to 'make a transition to un-real numbers' and do the math, the results are conform the observed reality. If you come to think about this, it is very counter-intuitive.
Besides, keep in mind that this WIck contraction (ie analitical continuation of the gaussian integral in the path-integral formalism) works practically everywhere in QFT !
regards
marlon