A History of Modern 64-bit Computing (2007) [pdf]
courses.cs.washington.edu
courses.cs.washington.edu
Very interesting read still.
Even then, the markets Microsoft was selling NT into, wouldn't have benefitted much from the extra memory (or the greater floating point performance) allowed by 64-Bit processors. Microsoft has never been in the scientific computing part of the business that SUN, DEC, and SGI were playing in.
Here's a classic post from John Mashey outlining the rationale for 64-bit in the early 90s: http://yarchive.net/comp/64bit.html
Also remember how long and painful the 16-to-32 transition was on Microsoft's platforms. They had every reason to get their platform and running up many years before customers needed the 64-bit apps.
IA-64 was a mistake, it has hurt everyone who touched it, I believe it killed VMS (beyond the crazy way VMS is licensed) (and did no favors for HP-UX either.) - it is and was too hard to build a really efficient compiler for - thankfully microsofts porting effort to IA-64 got the codebase set up to go to AMD64.
In many ways though, as much as I'd like to see other microarchitectures, x86 is the UNIX of hardware, it's good enough for most applications, and in the long run, good always beats perfect.