What's making the MXDs grow so much? Good question, I wish we had clear answers. Part of it is likely user behavior, although it is unclear what all behaviors cause the problem, but another part is how the MXDs are structured (design structure) and how ArcGIS interacts with them. Geoprocessing history may play a part in some of the MXDs, but it isn't the dominant issue with the majority of the MXDs.
MXD bloat isn't an issue amongst GIS staff as much as the casual GIS user in our organization. There are numerous worst practices that casual users seem to develop over time, and some of them may contribute to bloat, but there are limits to what can be expected from educating users on best practices. The bottom line for us is that we have an issue, prevention has its limits, and so we need to explore ways of cleaning it up on the back end.
Opening up MXDs and trimming geoprocessing history will shrink some files to varying extents. What also works is just opening up MXDs and saving a copy using the ArcMap UI. The shortcoming of this approach is that it is manual. We are looking for means to automate cleaning it up for thousands or tens of thousands of files. Right now, either spawning docDefragmeter from Python or shell scripting it works, it is just clunkier and slower than we would like.
