I think the when we talk about Twitter, the definition of "reliability" is entirely different. When we talk about emails, we can say it's reliable as long as my email gets delivered one way or another. For example, if I send a mass email to 20 friends, I don't assume anything about when each of them will receive my message. I just think that it will get delivered in a reasonable amount of time.

However what Twitter brings to the table is this: It is safe to assume when I tweet, all my followers will be able to see it immediately. This is a huge difference not because of the delivery time, but because of the type of behavior it enables. Being able to assume that my message is delivered in realtime is very different from not being able to assume. And this is not a technical issue, so even if you build a "good enough" realtime system using decentralized architecture, it wouldn't be half as good as Twitter.