There is a huge difference when one looks at the basic requirements of Server 2003 and server 2008. 2008 is based upon the Vista design which has a huge additional OS make-over attempting to solve some of MS' biggest past issues and footprint a new destiny that has moved us into the Win 7 era. But all that glitters, is not always made of gold.
My company markets VPS servers for dedicated users doing Currency or Forex trading, so the audience is fairly captive among the needs to be satisfied.
I have clients demanding Server 2008 R2 without realizing it comes out of the box eating up 400 meg of RAM, due to massive service dependencies that simply the core installation requires. Compare that to the Server 2003 requirements of just 104 Meg of RAM. THAT'S A HUGE DIFFERENCE in a 1 Gig RAM provision. You would think dropping all the unwanted services would help here, but unfortunately MS has interwoven unrelated services into the dependency map to keep you from doing just that. By forcing us to maintain all the services, the unskilled user is much less likely to break some functionality he might need in the future, so MS is forcing us to keep all the brain dead supports alive, to reduce the negative support exposures that come from optimizing performance.
Many have become fond of saying "RAM is cheap these days". But is it? Not when it's part of a leased package a user will re-pay for month after month, especially given it's to support a ton of unused services he cannot turn off, without losing some part of core functionality.
As 2003 is already becoming part of forced obsolescence, the day will come that I cannot offer clients the kind of memory vs performance benefit 2003 provides, which forces me out of the benefit I get by undercutting competition that doesn't understand this. MS continues to make efforts to ward off software piracy to a great extent required because of the policies that have evolved in rushing incomplete products to market, using the customer as a paying beta tester, then relying on patch after patch to finish engineering in a broken and inefficient process of wrapping up a design, just in time to label it obsolete and start over again. Must be some kind of economics at work here I can't understand fully, but it surely doesn't benefit the end user footing the bills.