Seriously? I understand the concept, but if it relies on 3rd party software/hardware for the user that has a camera to watch the user, it would never work.
Good concept, but bad implementation.
The only way that this would be remotely feasible is if you were able to remotely turn on a user's camera - which won't happen and wouldn't be taken lightly by the end-users.
Ofcourse - all of this doesn't include the cost of awareness that it would take just to make remote use of eye-tracking. You say "tens of thousands" of eye trackers are on the market. With 7billion+ people in the U.S., it's highly likely that you wouldn't receive one person using an eye tracker in the next 10 years - in which you would need hundreds just to make a reasonable estimate on potential site adjustments.
I get what you are trying to do, but the market just isn't there. A better way would be user predictions - or predicting user actions and engagements. That technology is around but not heavily used yet in analytics and would be able to make similar predictions.