Not to distrust Intel's study on temperature, but it makes a lot more sense for Microsoft and Google to raise their temperatures than smaller operations (like my company). We have a small server room (3 racks, mostly full), but we don't have seamless failover and often we're without hot spares. So, it matters that servers stay up.
With Google, it's not that big a deal. They expect failure and so it simply becomes a calculation of what's worth more, the electricity or the hardware and I'm guessing the electricity would win out in that situation. However, for us it's the hardware and the downtime and the hassle of setting up a box to support some ancient system some of which, no joke, date back to DOS days (or maybe I'm just really young and think that's old). Suffice it to say, it isn't seamless and it isn't even as easy as installing and configing a LAMP server.
I have been there and done that. One thing you should at least consider is seeing if you can get some of your more creaky servers to run in a VM, you can run DOS in virtualbox and probably xen as well, you might be able to replace a number of physical machines with a single virtualisation box ( a pair if you are serious about availability ).
Going this route should save you energy and space. If you can replace a rack full of machinery with 1U of server and 2U of disk shelf, you are way ahead of the game.
With Google, it's not that big a deal. They expect failure and so it simply becomes a calculation of what's worth more, the electricity or the hardware and I'm guessing the electricity would win out in that situation. However, for us it's the hardware and the downtime and the hassle of setting up a box to support some ancient system some of which, no joke, date back to DOS days (or maybe I'm just really young and think that's old). Suffice it to say, it isn't seamless and it isn't even as easy as installing and configing a LAMP server.