OK, I know it's bad form to answer one's own post...twice...but I stumbled upon a working solution.

As I said, using the codes learned on the TSU1000 seemed to work better than the database codes, in that the volume jumps were 2 db rather than randomly 2 or 3. This was from testing using the "test IR" function in ProntoEdit. So, I decided to upload those codes to the 9600. The durations were set to 0.1 sec.

Surprisingly, even though "test IR" gives 2 db jumps with these codes and these durations, once they're uploaded to the 9600, they work as intended--1 db jumps and continuous change if the buttons are held down.

I guess the moral of the story is don't assume the behavior through ProntoEdit is the same as with the Pronto itself. And, of course, Lyndel was correct in his advice.