I don't think the lack of NKRO is a factor here - the keyboard is still capable of tracking lots of keypresses as long as they don't trigger the anti-ghosting or whatever, and in most cases that wouldn't light the Teensy's LED.
I think this is the keyboard's buffer filling up, at which point it sends a code saying that it has overflowed. The converter then clears the state of all keys and lights its LED for a 5 seconds. In nearly all cases where the LED lights up, the converter solves the problem so you don't notice anything bad happen :-)
I've noticed this a lot while testing, particularly when releasing many keys, and there's two reasons for that: first, you are more likely to release them all within a very short time than when pressing them, and second, for scan code sets 2 and 3, there are at least two bytes placed into the buffer for each key release, making much more likely to overflow the buffer.
As for that 122, I'm afraid I have no idea

It could be a different protocol, or possibly the connector just isn't wired the same as a normal XT/AT board, or both.