Are you ignoring the 2 recent non-MS TCO studies then?
I really have no desire to get into a TCO argument, but there are two things that I always, always find funny when these studies come up:
1. While a company doing it's own study does not have the appearance of partiality, that does not mean it is biased. By studying the methodologies, test subjects, etc (which Microsoft hasn't lied about, that I'm aware of, in past studies) you can easily determine how indiciative a study is (most people simply would rather say "MS did it themselves, or they paid for it so it doesn't count")
2. Where studies are done and conclusive evidence is found, they are ignored.
Personally I don't care about studies. Whenever you do a conversion project you should run your own numbers and those numbers should be one (just one) of the deciding factors.
Either way we are geting severely off track. This was a discussion about what could be done to improve Windows in the server space.
As far as Windows 2003 and performance (getting back to the subject at hand), most servers in companies our size are single-purpose, single-app servers (about 50%). We generally don't buy with power in mind.
For instance we just bought 25 Dell PowerEdge 2650's (2 x 3.0GHz Xeons with 4GB of RAM and a pair of RAID arrays (one for OS and one for data)).
None of these will ever max out their power requirements. As such, while the added speed of 2003 is good to know about it doesn't help us.
What will help us is ways to bring down the cost of SAN-attaching servers, a more streamlined way to interoperate with other tools (particularly Novell ones as they are the most mature management tools (in my opinion) outside of Microsoft) and better asset management built into the servers (ie: outside of the DSI management tools).
Disclaimer: again, we haven't even got our 2003 test lab up yet, so there could be significant improvements in these areas already.