I know it's a long way from how it might play out in the server space, but I've had this personal experience with a Dell Mini 9 netbook:

The machine came from Dell with Ubuntu installed. At one point I thought I'd try Windows XP on it, so I nLited a new XP disc and installed it, using advice I found in a tech mag. Ubuntu had booted in about 40s, so I was disappointed to see XP take over two minutes (I define "boot" as the time from button press to full functionality, including WiFi). I tried another nLited install, with identical results, then a full install. Of course it's wimpy hardware, but it was just unacceptably slow compared to Ubuntu.

So then, just for fun, and because I hadn't been fully satisfied with either Ubuntu or XP, I tried "hackintoshing" the little beast. I put Leopard on without problem, using instructions posted in the Dell Mini support forum. Result? It transformed the computer. Start is back down to 45s, applications open almost immediately, and it's quite acceptably fast - feels like a normal good computer.

The revealing thing about all this is that all three OSs were running on identical hardware. It's obvious that OS X was making much better use of that hardware than XP and slightly better use of it than Ubuntu.

The Leopard install is the full OS, with MS Office and other typical productivity apps. You'd think it had far more horsepower under the hood than a single-core Atom 270. I think that's a pretty impressive demonstration of OS X's overall efficiency.