Is there any practice of using biometric mechanisms online - for example, for visual identification via webcam to identify audio-visually the person speaking, or, performing biometric analysis on keyboard input? Particularly flash web control would be useful.
Yes - both are current.
Real time biometric analysis on keyboard input has been looked into by a few teams now. The guys at Pace University seem to have a strong presence in this area. Have a read of :
The first one is probably more what you are after:
As far as video goes, all the facial recognition programs used in identifying users in public places work off video streams, either using single frames, or collating multiple frames to improve quality - this is used all over the world.