So why don't they do something like: deprioritize traffic in proportion to how much you that user uses at peak times. (Shouldn't be hard to turn it into a more rigorous formula.)

So the pipe is normally shared equally between requests, but a heavy user might get only half a normal share. That way, they optimize against the real scarcity (in peak time usage, not total downloaded) but no one has to worry about overage charges, and any throttling is to the benefit of other users.