In this case the PSU is capable of supplying a lot more power than a fully fitted out server can consume at maximum load, and then some. If the server doesn't have a dozen disks it's likely to still have the same model of PSU as the one that does, and even that one with a dozen disks is not going to be running them all at maximum power consumption all of the time. For example I've got compute nodes with a single SSD and a dozen empty bays because that's the cheapest decent chassis for a multi-socket board - the max load on those PSUs would be a small fraction of wjhat they can do.
The real answer to this is just look at the power bill to see usage or run a server via one of those now really cheap meters for a while and see what it really draws.
Apart from outliers like the Pentium IV (netburst) stuff, and storage of course, the power consumption of typical machines under typical loads hasn't dropped as much as we'd all like over the past few years. That old Sun machine sitting idle most of the time is not going to use much less power than a new Xeon sitting idle most of the time.
If you go beyond the typical and use an Atom or ARM chip and SSDs you win big time (if your software is portable enough) but comparing like to like you don't.