ralph27 Spinebot's answer is not wrong, but I would like to emphasize that it is the number of vertices and the complexity of the weight settings that should be of concern, not the number of meshes.
For example, if the skeleton has 100 mesh attachments, but each has 4 vertices, it is not that complicated. The performance cost should be even lower, especially if some of them are used to change the appearance. For example, if 20 of the 100 mesh attachments are for facial expressions, such as eyes and mouth, and only one of them is shown, this is actually a fairly simple skeleton, since only the active attachments affect performance.
If the skeleton has only 10 mesh attachments, but each has more than 100 vertices, it is more complex than the one above. Also, if each mesh is bound by only 2 or 3 bones, it is not that complicated, but if it is bound by more than 10 bones and each vertex has a weight of all of them, it is much more complicated because it requires many more calculations for the position of each vertex.
For additional information, see the Metrics page as recommended by Spinebot.
