Bizarrely, this whole mess got figured out when the Boss showed up. It turns out that the guy who wrote the code I had to talk to took his original skeleton code from said Boss, who probably had a multiprocessor application in mind. That got propagated into the guy's code, which is now in use throughout most of our products. And of course nobody spotted it until I ran into it headlong.

Ah well, it's been a learning experience. Many thanks to everyone who helped out.

Oh, yes, about the actual problem: the device I was talking to was set to use the ninth data bit, and the TB8 bit was set high, so that it sent what seemed like a second stop bit. I guess I toggled the wrong bit when I tried it earlier. Sorry about all that.

Many Thanks,
Bob Robertson