A model I'd like to see attempted - 2 hours of lecture (video) + 1 hour of recitation (human supervised), in lieu of 3 hours of lecture+recitation by a human. Now a single human can teach 3x more classes. That's a massive reduction in costs, but I don't see how it can significantly increase educational outcomes.
There is one caveat - I expect lecture quality would improve a bit. If only the best teachers record videos, then lecture quality will be better than average. By definition, human-delivered lectures can only be average.
I really like Khan's model of doing lectures/learning at home -- and using the classroom time for exercises // with the teacher (or other students) mentoring those who are stuck on a problem set.
It's unclear to me that even a 1/3 reduction in lecture time is sufficient. I'd like to fundamentally change education so the teachers are really mentors and facilitators of learning and activities. Lecturing can be done via video almost exclusively -- at least for those topics where video lectures make sense.
from your post above:
>and a 16% increase in efficiency (using 2007 numbers) would be sufficient to balance the federal budget deficit [1].
that means anywhere between cutting 16% of cost while maintaining the same educational result and increasing the educational result by 16% while maintaining the same cost. I'd argue that in modern world (i mean today and the next 30 years at least) increasing the educational result would be much more important than cutting the cost.