Well, we officially don't support this, but the data is in one of the "other" FLV files (containing data messages). Not sure of which one, on the top of my head (but I am guessing the name is related to your screensharing object id) but the content should be a series of calls to publishItem with the mouse coordinates and more stuff.
You'll need an FLV parser to extract the data but again, it should be just a bunch of calls at the right timestamp (that you'll have somehow "mix" back with the video stream). If you make any progress on this please let us know. I am sure other developers would be interested.