Good on him for trying! This'll surely be a failure of ambition, if it is one.

Q) Why not any assignments, papers or projects?

Simply put, because self-grading those things would be almost impossible. I chose final exams as the basis of evaluation because, for most classes I’ll be following, the final exam is a good basis of evaluation, and because it is relatively easy to be objective when grading.

The impression I get from most comp sci students (and the CS classes I've taken) is that the assignments are where a lot of the learning happens. If the professor of a course had most of the points coming from assignments (I do not know if this is the case for any MIT CS classes), then that suggests to me that just taking the final is an insufficient measure of proficiency.

Likewise, his serial method of "consume a class, test, repeat" seems to fly in the face of spaced repetition. If he manages to get a D in the final exam for all the classes a week after reading through the lecture notes, but then does poorly when he takes the first exam again, this suggests this may not be a useful method to learn material. (Of course, there's good reason to believe the semester model also sucks.)