I posted my hardware info to you in the FreeBSD-hackers mailing list (zander).
My AGP aperture size is the commonly recommended half the size of my main memory, so 128 MB. I was looking closer at the code, and the comment after rm_alloc_agp_pages in nv_alloc_pages says that rm_alloc_agp_pages will only fail if I am out of agp aperture (memory?), so I think increasing my agp aperture might also make this bug either go away or occur less... although it's not a real fix, just a workaround.
You're assuming right in assuming that I'm using -STABLE and that I've compiled everything without optimizations.. etc...
And you're right... all that's necessary to trigger this problem is for me to run UT2003. It never even gets past the splash screen.