One of the most consistently frustrating aspects of Web3D is telling people to go to some cool new web site and qualifying it with the ..."but you have to install the 3Dxyz plug-in and it really only works well with Internet Explorer"...excuse. Plug-ins are not the worst thing in the world but they don't help and are often a signigicant impediment. I'm also the first to point out that FLASH requires a plug-in and has certainly been a major success. So the existance of a plug-in alone is not the only barrier to technology adoption but it is a factor. You can find lots for FLASH info at Sue Chastain's Graphics Software site.
The recent work led by the Web3D Consortium on a next generation VRML called X3D has resulted in two significant pure Java implementations which are really VRML-light. Shout3D and blaxxun3D both offer quite good renderings and speed with NO PLUG-IN. A more thorough evaluation was presented a few months ago. It is SOOOOOOOOOO GREAT telling someone to go to a site and knowing it will just work.
The big advantage of these systems is simply that the 3D on the page just appears. No plug-in, no download and installation instructions. Java takes care of all that nasty code because it's already installed. Another added bonus is that Java (with minor tweaks) really does work on multiple platforms including the oft' neglected Macintosh.
The bottom line is if you're developing Web3D, unless your doing work requiring the specialized capabilities offerered by proprietary Web3D systems, you really should see if your project can be accomplished with one of these pure Java solutions. If you're a viewer of 3D just relax and go to the sites that use these technologies.
Another pure Java renderer is Anfy3D, supposedly the origins of the blaxxun3d work. There is also the Janet3D system another pure Java renderer. Their approach is not to offer a general 3D solution but to provide additional infrastructure via a server which you must use to take advantage of their 3D, so it doesn't appear to be a ffeasiblegeneral purpose renderer for random building.
Simply put these pure Java solutions offer a sufficient performance and fflexibilityto deserve a serious look. In particular if you are concerned about obtaining the widest possibly audience for your 3D content.
Of course the pure Java 3D rendering systems are somewhat light weight. They don't offer the full power and flexability of VRML97. What do you think about that?