How were the results calculated? Did the listener press a button or similar? IE: Were you mesuring reaction-time? To my knowledge, reaction time has been measured to be in the 200-300ms range. Which is quite high.
But reaction time is not quite what this issue is about. Humans clearly can recognise audio latency at lower levels than that, otherwise the chorus effect wouldn't work for our auditory system I would think??
Anecdotally, I know I cannot deal with anything higher than 10ms roundtrip when playing midi keyboards and drums. For guitar it's a little lower, 8ms round-trip.
As I recall, voice calls have a target of something like 150ms latency. But for gaming applications they aim for 13ms round-trip.
It seems our tolerance for latency is heavily impacted upon the level of interactivity we have with the system. Passive interaction can be much looser, but actively interacting would seem to require much tighter timings.
Interesting field, wish I had more time to properly study it.