I probably don't have the answer you were looking for, but my limited understanding is that bow-tie cavities are primarily used for optical parametric oscillators (OPO), where two mirrors will be curved to increase intensity in the nonlinear crystal. In this case, you don't want to make the curved mirrors partially reflecting, otherwise you won't get mode matching. That leaves you with two equally good choices (the non-curved mirrors), and in enclosed lasers the choice is usually made by the physical layout.
For a generic bow-tie cavity with all planar mirrors, I don't think it makes much difference unless the arrangement of mirrors is highly asymmetric. If you were determined to find out, the highly asymmetric case could be solved by an ugly and arduous application of the Fresnel equations to each mirror. To be real here, it would be faster to test this empirically on a breadboard than it would be to prove a priori. What you would be looking for is at which positions the output coupler is least likely to support parasitic oscillations (internal reflections in the coupler).
Sorry if this info is too specific to certain applications. Can you tell us more about what the bow-tie is for? That might allow us to provide info better suited to your needs