I really like three.js, I use it, and I like that people are doing good things like this with it.

But...

I wonder if someone was building a 3D library for the web today if they'd do things the same way as three.js does things. Under the bonnet three.js is effectively a bunch of string concatenations that build up GLSL shaders for things like materials. It works well, but if you look through the code it really is a massive series of if statements (and loops? :) ) like "if a lambert material else if phong else if standard else if physical, and if lights > 0, and if env maps.." Hundreds of them, all smooshed together. The external API that you use to access everything is quite clear and easy, but doing things extending a material is much harder than it needs to be (you essentially use a regex to modify the big concatenated GLSL shader string..).

I can understand wanting to use the API in another language, but I'm less sure about wanting to use the internals.