I even had a mate who made a navigable 3D tour of his house in it. Worked reasonably well even on the scrappy hardware available (if by "well", you're satisfied with output sitting somewhere between 3D Construction Kit on the 16-bits and Quake on a non-3Dfx'ed Pentium) but must have been a bitch to program because he lost patience with completing it. Can't remember if it actually did need a plugin... surely it would have been integrated into browsers as time went on? It's not that complex an idea really. If you can fit complete 3D games for older systems into less than a meg, then putting one in a hijacked image frame in a browser can't be too heavy on the resources?
Plus somewhere in my inherited collection of cheap tat CDs my dad couldn't stop buying from the local computer shop, there's a "3D Web Designer" program that ostensibly spits out working VRML sites with an "easy" bit of mousework. I'll have to fire it up, see if it works, see if browsers still accept it.
What in all this suggests that WebGL won't go the same way, anyhow? I mean, if the only thing that was stopping VRML was a "lack" of accelerator cards (they weren't THAT rare at the time, really, certainly not for people who could afford a modem and by-the-minute internet access...) or decent CPUs, crap bandwidth and hooky plugins, all of that could have been ironed out or just improved by the passage of time for VRML. And people's higher standards for online content these days (games like Crysis, flash-heavy sites and video streaming, stuff that actually does tend to "just work") may mean a similarly critical eye is cast on it even though it could be piles better. Could just be that 3D interfaces actually don't work too well if you're not physically walking through them? I mean, we do have Second Life, but that's not useful for a great deal of "real" use, other than the odd gimmicky virtual seminar.
I say put WebGL to simmer on the back burner for a bit until we have fully immersive cerebral hookups to our PCs, then its time will properly come.