This is a great observation. Part of the reason for this behavior is almost all the people I've met who tout usability tests come from the old school market research/focus group approaches. Put people in a room and listen to what they talk. I've experienced this first hand. I've even put up mock print screens of the click flow and inevitably there is a disjoint. That does not work - the only reliable way is to put the actual working app out there and observe them use it. OBSERVE the users (with the UI guy in room) and correlate/validate what you see with objective page abandonment/scenario data on Google Analytics. Its a little bit of work but I think it totally completes the picture. Excellent observations.