There are a few really big pieces missing from this analysis.
1) Orkut is a .NET application and the rest of their apps, as far as I know, run on customized linux. So, unless Orkut is entirely separate (which seems unlikely), either they've built a cross-platform cluster or they're running Orkut on mono or something like it. Either way, this is still more powerful than the article above would lead you to believe.
2) It sort of dances around what I'm going to say next, but fails to make the next conclusive leap. The Google cluster is supposedly based on the assumption that hardware is cheap and disposable - it routes around damage. But what if, from Google's perspective, hardware was not just effectively free and disposable, but >ACTUALLY FREE< and disposable? I suspect that they've built or are building what I've been talking about recently - a way to use the entire internet as distributed storage. Because of the Google toolbar and the ability to correlate activity to IP addresses, they've probably got a pretty good map of uptime for a large chunk of the internet (which is in itself, extremely valuable). Why should they pay for a data center cluster to run your Gmail (or whatever), when they could make you pay for it instead? To a system like this, disconnecting a machine from the internet or turning it off looks exactly like a failed disk. Sure, it happens more often, but you've got plenty of space to spread things around in for multiple layers of redundancy.