Depends what you mean by "server". Is the service you're interested in available only via that one server? Specifically, you could host your own twitter-equivalent handle from your own server, and the server going down wouldn't bring down the twitter-equivalent-service. Or your server with your blog going down doesn't bring down everyone's blog.

I'd say we should keep the decentralised and replicated-for-reliability concepts slightly separate. They are interesting together, but they're orthogonal.