Yea any intelligent bill limiting needs to be done with context into what each instance is doing. It's possible for them to implement something that can handle that, using tags and such, but I would rather they spent that time improving reliability infrastructure.
It's not terribly difficult to implement one's own bill limiting system using the API.
It seems impossible for a single solution here. One business would rather see instances go down, another might need ledger logs no matter what and would rather turn down customers at the door, etc.,
The right way to do it (currently) is to model the costs yourself and tie it in with your operational automation, then determine what your best course of action would be during a shitstorm. Generally, I would wager the best option is to replace your front page with a static html "overwhelming response" page, maybe with some light vector graphics.
It's not terribly difficult to implement one's own bill limiting system using the API.