5.9 years ago
Edit: Sorry to necro this post, but I've had old things pop under my open list from time to time. I don't always notice the date. My bad.
So, what exactly do you have calibrations for?
If you are placing a prior distribution on the age of a node, and the node is well-resolved with your current dataset, it may not be necessary to add a bunch of additional sequences. Think of this as trying to resolve family-level ages and you have representative individuals from the genus-level. In other words, adding more taxa descendant from this node will probably not have a huge impact on the age of the node itself (though many factors could affect this).
If you have calibrations for the substitution rate, adding more data will help improve (i.e., narrow) the 95% HPD of your posterior up to a point. This is assuming that you're using your knowledge about the rates to set a reasonable prior and letting your data guide your posterior estimates.
Both of these depend on the information content of the additional sequences that you're adding. If you are truly fixing (i.e., not allowing values to change as the MCMC progresses), you are estimating your distribution of trees with those constraints imposed. If your additional data supports those constraints, the likelihoods will improve, but there may not be significant differences in parameter estimates. If you have the data handy, I'd try both and see.
Personally, I would take a total-evidence approach and use all information available to be as confident as possible about any conclusions.