The serialization of the skeleton was a new development with the release of the official Kinect for Windows SDK v1. While the format is not ideal (text as opposed to XML, due to inherent limitations in the skeleton-joint object model), neither is the size, which is something we’re working on.
If you are producing up to 30 frames per second, you are going to create up to 30 binary serialized object, or recordsets, of the skeleton information. Saving these out to text right now (as a flat file) produced a very predictable 2.2 KB file. If you are recording this activity over a significant period of time…well, I will let you do the math. Thus, the baseline serialization is not optimal for this particular application.
Without losing any of the good stuff that is included in this serialized object for playback and analysis, there are a host of other options to chose from. The current data set would rate equivalent to an old ISDN line @ approximately 64 kb/second. And, remember, this is NOT including the heavier information, like color and depth images to video. So, this is certainly something to consider when capturing this information.