I know that's the conventional top down narrative and I think you presented it well.
I'm proposing that as a bottom up narrative, they continue to include extensive grind barriers such as peculiar scheduling and pacing, someone else designing arbitrary barriers in terms of assignments with firm due dates, etc. Yet the reward has been cut out as per reasons in the top down narrative, etc. So from the bottom up perspective you're left with a stereotypical grind game where the reward has been removed, and from that theoretical model, the dropout rate makes sense. Then when people sign up and get something out of partial participation, it makes no sense to use a successful grind game metric measurement when its a failed grind game therefore the players aren't cooperating because they feel no need to do so. Because "MOOC as a grind game" has failed, a MOOC is a place to learn, not a place to jump thru hoops or not get a reward other than making someone else's completion metric result look good.
You can run a grind game without a vocational meal ticket reward model. WoW and Eve do just fine.
My suggestion would be to use advanced technology to abandon the grind game aspects, assuming higher completion rates are inherently valuable. Even if completion rates are not valuable, removal of pointless grind will result in better operating conditions for the students. Why must class schedules correspond vaguely to the northern hemisphere agricultural growing season? If you're not teaching project management, why are due dates so important? Why is there (typically) only one very fixed learning track if more exist or there is a "chose two from column B and three from column C" aspect inherent in the field itself? All of that would be very difficult to implement at Harvard in 1640, but due to technological advances would not be very challenging to implement today.