If one server goes down and they are not able to restore it remotely they will just write it off. If a large number go down due to some common failure, it might make sense to haul the whole thing up and refurbish it.
I've heard that this is what many data centers currently do anyhow. That when a client "moves on" (moves out, dissolves, etc) - the machines are just left behind, because it costs more to unrack/transport/etc than it does to leave them in place.
/source: worked for a small cloud computing VPS provider
