Yeah, IIRC, it was the first DAS that had transposition errors. People thought it was because of a low scanning rate, but when measured it wasn't particularly low. My theory is that it had a poor debounce algorithm - batch debounce can group keys up, and usually they're then output in scancode or HID code order (for PS/2 or 6KRO), or the order they appear in the report (for USB NKRO).
Pressing keys simultaneously doesn't really give any useful info for debugging the issue, since that's (hopefully) not how you type! A slightly better test is to bring a straight edge down on a few keys at an angle, so you'd expect them to come out in a certain order.
I don't know much about the KBT Oni though, so can't really say what might be causing the transpositions. The NKRO over USB isn't really needed for typing (an NKRO matrix using 6KRO USB should be sufficient), but on the other hand shouldn't be causing this problem.
Technically, the features needed for fast typing are a high scanning rate and a decent debounce. Preferably individual debounce for each key, although batch debounce might be sufficient with a fast scan rate (say, 1000Hz) and tight timing (say, 5ms). Trouble is, manufacturers are not particularly open or clear about these specifications. So, currently, I think the only way to be absolutely sure of what you're getting is to go with a custom controller, for example one of bpiphany's or a Teensy with hasu's code (or my code when I release the controller version generally).