Independent comparsions have been done - here are two extracts from comparing software to fit a mixed effcets logit model:
http://www.biomedcentral.com/1471-2288/11/77
"The speed of the Bayesian procedures appears to depend also more on the sample size than the frequentist approaches.As a result, long processing times as in WinBUGS (14 minutes for binary and 8 hours for ordinal model, respectively) may prevent the user to do much on exploratory statistical research. The R package MCMCglmm and MLwiN (MCMC) were much faster than WinBUGS, taking only a few minutes for both binary and ordinal cases. Hence, from a computational point of view, MCMCglmm and MLwiN (MCMC) are our software of choice for multilevel modeling. In our experience, the SAS procedure MCMC was inefficient in dealing with mixed models. It was far too time consuming (37 hours for the binary model) and it did converge neither for the regression coefficients nor for the variance of the random effects. At this moment, we cannot recommend this SAS procedure for fitting logistic random effects regression models."
and
"For a Bayesian implementation, we would recommend MLwiN (MCMC) because of its efficiency"
In full disclosure mode I should say that I work at the Bristol Centre for Multilevel Modelling that produces MLwiN but that I do not profit from sales.
I aslo realize thatthis may not help with phylogenic searches!