Yeah, the only real question is whether or not it's true that they no longer do intelligent routing. If that is the case, then regardless of anything else the problem exists once you pass a certain scale/request cost. It won't matter if that one dyno can handle hundreds of requests at once, it will still queue stupidly.