Thanks for your comments. Yes, Teneyes, the bus setup and the trigger setup are good. My 115200 baud tests use a 1000-byte stream, output at full rate i.e. there is only approx 2 usec from the end of a stop bit to the beginning of the next start bit. The waveform quality and timing is excellent. I have tested different memory sizes, starting with a fast time/div setting, and repeating with slower time/div until it fails. Here are simple results:
Memory = 1.4MB
Works at 2msec/div, 5msec/div (probably also at shorter rates).
Fails at 10, 20, 50, ... ms/div
Memory = 14MB
Works at 2msec/div, 5msec/div (probably also at shorter rates).
Fails at 10, 20, 50, ... ms/div
Memory = 140MB
Works at 2msec/div, 5msec/div, 10msec/div, 20msec/div (probably also at shorter rates).
Fails at 50, 100 ... ms/div
Obviously, as the time/div goes up, the sampling rate decreases, for a given memory size. But the sampling rates at which it works and at which it fails are not consistent across the different memory sizes. I really can't see the logic behind the failures.
Maybe it's something to do with their algorithm. Certainly its fast, only taking 2-3 seconds (although panning back and forward is painful). My Tektronix MSO4034 takes nearly a minute to decode 10MB of memory, but it works every time, every setting, no problem.
Another gripe: Trying to set a USER baud rate is a lost cause on this scope. The numbers change nice and smoothly for a while, then suddenly jump wildly, by hundreds-to-thousands of counts. Turning the knob backwards a little (if the numbers had jumped forward) will often produce a jump back to near the value that you were at when the first jump occurred. Or not!