How to decide which model is "best" when they differ by priors

[Jamie T]

My analyses seem sensitive to the expected number of rate shifts, with the RTT and BSC changing quite a lot between models.  It is a large tree with >5,000 species, but all analyses (5, 20, 50, 100, 150, 250 shifts) have reached convergence, with an average of 150 core shifts.  Is there a way to decide which model is best?  And is it as simple as looking at likelihood?