There were dual socket x86 motherboards around when NT4 came out? I have a quad slot2 P3 from 1998, people have mentioned dual PentiumPro's etc, so SMP was very much alive and kicking when NT4 came out.
Does any motherboard exist out there with more than 4 sockets? doubt it, that would be a very big motherboard, no point when you can utilise space perpendicular (i.e daughter cards to hold any extra circuitry/silicon/CPUage)? Love to see a pic if this is the case though. o.0
Multi CPU NT4 would have been marketed mainly for the Alphas imo, primarily due to the fact that, back then, you could count the number of programs commercially available on x86 that utilised SMP on two hands probably. Maya batch rendering was the only one I knew of (version 2.5, prob went earlier but that was the first version I used), there were no doubt others. While the instruction set of the compiled kernels and programs is different, both Alpha and x86 variants would utilise the same 'NT' technologies, so in theory an x86 version could support 8 CPU's, just like it's RISC sibling. There just weren't many of these systems about... too niche, and if you did find one, I doubt MS would have actually tested on an 8 CPU x86 system given the limited market for it, so stability might be an issue. Alphas on the other hand, are a different story, many Alphas with many CPUS around in 1997 so hopefully would have been widely tested.
Also NT4 is an OS for a shared memory architecture (parallel), not distributed, so blades and most clusters won't work with it. you need Windows CCS or HPC for clusters.
[EDIT:] correction, the Quad slot2 P3, is what I upgraded it to. It would have been Quad P2/450 when it came out in 1998. Apologies, my mem appears to be corrupted.