Saving on power in the datacenter is one of the worst ways to try to save money, in my experience.
It varies greatly, but $10 million dollars worth of servers, network gear, and storage might cost $25k/mo in power. Shaving 5% or 10% off that bill is practically meaningless. You can save a lot more money in other ways.
Obviously at massive scale (think Google) it's a different story, but for the average small-medium company it's not even worth thinking about until you've gone after everything else.
While I agree that for those numbers, 5% to 10% savings isn't worth the effort, this isn't a good reason to dismiss saving by reducing data center power. Nissan recently starting rolling out virtualization solutions for some of their datacenters, resulting in 34% energy savings. For the kind of numbers you list, that would end up being over $100k of savings a year.
It proves exactly my point: That focusing on power usage is backwards. You should worry about other stuff, like using less servers.
The guy that worries about using less servers can consolidate 159 servers to 28. He saves $5 million dollars and $10k/mo. The guy that worries about power usage itself just buys 159 slightly less power hungry servers. He saves $2k/mo.
Ah, now I understand the distinction you are making. And I agree; get fewer servers, not the same number of slightly more efficient ones, and then you can talk about saving money on power.
Our colo was built back when computing per watt took up a lot more space... meaning we had two racks of machines in a giant empty room and we'd already blown our power budget.
It varies greatly, but $10 million dollars worth of servers, network gear, and storage might cost $25k/mo in power. Shaving 5% or 10% off that bill is practically meaningless. You can save a lot more money in other ways.
Obviously at massive scale (think Google) it's a different story, but for the average small-medium company it's not even worth thinking about until you've gone after everything else.