I've never written a "twitter clone" but I would suspect the real problem isn't pure throughput but rather the mixture of having to n-way route messages to the proper recipients combined with having to spread the load of that across potentially many systems. I do not think doing that is as trivial as it may seem.

However, I'm sure you're right in that someone could reimplement what twitter does more efficiently than they do it if they paid proper attention to all the mistakes twitter made along the way. Not sure it would be worth the bother though, because twitter's real win is having their company name define the action in much the way googling has come to define web search. You can't overcome that by being marginally better.. or, as it happens, even by being marginally better and giving away tons of free money (just ask MS/Bing).