This is going to be difficult since most 3D CAD programs do not take into account the possibility of revision, so when you load something and then save it again it may completely re-order the points (there are reasons for this, usually done for performance).
Further, large models represented in a text format are huge files, and will take forever to copy/merge/etc.
There is no current system that will manage this, but there's a really big need in the industry for it.
I would expect such a system would have a model normalizer that converts to and from the desired CAD format and the revision format. It could then handle merges and track changes more easily.
It would also need to output diffs in a form that you could open a "diffed" model in a cad program and the changes are shown in a different color or otherwise highlighted. No one is going to be able to look at a text diff and understand what they're looking at. This diffing program would ultimately need to support understanding that two models are the same even though the 0,0,0 location and rotation are not the same (difficult matching problem) and give the user some interface to allow them to help it when it gets stuck.
You'd probably have to deal with the parts of the model separately (bones, mesh, textures, etc) and have a third file that synchronizes them when converting them to an inclusive model file for use and modification.
It's not a trivial problem... But if you started on something that just handled meshes and open sourced it, you'd probably get a lot of people interested.