Pushing issues upstream does not make them go away, it makes them someone elses problem. If that someone else doesn't take care of those issues then you end up having to solve them yourself.
And I've played enough with software raid that I know that configuring it to perform well is not a walk in the park, in fact I think it may be harder than getting a good hardware raid solution up and running on a dedicated box.
The nice thing of course of an EC2 setup is that once you've figured out how to do it you can do it again without much trouble.
As for the energy usage and the physical plant, that is entirely up to your way of using your servers. For instance I try to balance the quality of the service with the load on the servers to gracefully degrade the service when the load is at its peak (which is only a few hours per day anyway). That way I maximize my flat-rate payments on bandwidth and serverlease at a relatively small fraction of what it would cost me to get similar performance out of the various cloud suppliers offering. In a cloud environment those servers would be using just as much power and AC as they do today.
It takes me a little longer to get a server provisioned, on the order of 2 to 3 days, but that is 2 to 3 days, whether I order one, 10 or 100 servers. Very few businesses would ever need to grow faster than that.
Maybe my business is a lucky one in that it can make optimal use of a dedicated server setup but I see plenty of people choosing for a cloud based solution when if you run the numbers it makes very little sense.
The cloud comes in to its own if you have wildly fluctuating loads and/or jobs that need large numbers of machines for a relatively short period.
But for the majority of longer term high bandwidth uses I can't make the numbers work at all.