Probably the most important benefit of Bayesian analysis regarding unsupervised learning is the ability to compare models. Not only can the different ICA-models be compared to each other, they can also be compared to vector quantisation or any other statistical models, provided that Bayesian analysis is also applied to these other models.
In ensemble learning, the treatment for missing values is very simple. Since they are unknown variables, they are included in the ensemble and therefore their distribution will be estimated similarly as for any other unknown variables.