[By statistical mechanics I mean classical statistical mechanics throughout this answer. If you are curious to think about the complications added with making the statistical side of the story quantum mechanical, that sounds like a very good exercise. For clarification look at Chap. 3 of "Conformal Field Theory" by Di Francesco et al.]
The analogy between "Euclidean quantum field theories" and "equilibrium statistical mechanics near second order phase transitions" is exact, once you identify $\hbar$ (on the quantum side) with $1/\beta$ (on the statistical side). Being careful with the terms Euclidean and equilibrium is important to avoid misguided analogies. The proximity to a second order phase transition guarantees that (the continuum limit of the underlying statistical system approximates it well, and thus) the statistical mechanics can be well approximated by statistical `field theory'.
1) Roughly speaking, in real time quantum field theory, every intermediate stage happens with a probability proportional to $e^{iS/\hbar}$. Often you interpret those intermediate stages as "virtual particles". In Euclidean (or imaginary time) quantum field theory, there is no "intermediate" stage, so the right interpretation is (not in terms of virtual particles, but) that all possible classical configurations contribute to the partition function with a probability proportional to $e^{-S/\hbar}$. Now to connect this Euclidean QFT situation with one in equilibrium statistical mechanics near a 2nd order phase transition, one only needs to specify in what sense "all possible classical configurations contribute to the statistical partition function with a probability proportional to $e^{-\beta S}$". The sense in which the above statement is true in equilibrium statistical mechanics is of course, the Ergodic sense.
In sum, the answer to your first question is that i) the virtual particle interpretation does not apply to Euclidean QFT (which, unlike real time QFT, is analogous to equilibrium statistical mechanics near second order phase transitions), ii) in both Euclidean QFT and equilibrium statistical mechanics, every allowed classical configuration contributes to the partition function; it is just that in Euclidean QFT this has a fundamentally probabilistic interpretation, whereas in equilibrium statistical mechanics it has a statistical interpretation supported by the Ergodic theorem.
2) Yes. In fact, every Euclidean quantum field theory can be regarded as describing an equilibrium statistical physics system near a 2nd order phase transition. The term Statistical Field Theory is applied whenever the field theory is interpreted as describing some statistical system.
3) There is no Aharonov-Bohm effect (in the sense of electrons propagating and interfering with each other) in Euclidean QFT. This is a confusion similar to the one with "virtual particles" which is due to not keeping the word Euclidean in mind; there is no propagation in imaginary time QFT. Also on the equilibrium statistical mechanics side, there is no such a thing. However, if you are looking for manifestations of non-trivial gauge bundles, you can find such manifestations on both sides by looking at Wilson loops circulating around solenoids installed in your quantum or statistical system.