I've held off on making a post about this for a while now just to see what I could do with it, but I've made my own GLSL-esque fragment shader renderer on the CE. Obviously, the CE has no graphics acceleration hardware at all so the performance isn't going to be great, or even real-time. So far, I have written a few demos and attempted to port some ShaderToy shaders (which hasn't gone well). You can see some screenshots and download it on its Github page: https://github.com/dozer8383/ce-shader .
Right now I'm working on getting it running a little faster, but if y'all have any input on how to make it faster, that would be greatly appreciated.

I took a look at the code, and I see it's exactly the direct port of the concept one would hope for a starting point: for every pixel on the screen, a shader function is called with a floating-point value (0..1, 0..1). I can imagine trying to "accelerate" this by putting the shading in an interrupt, to at least decouple screen updates from the rest of the logic, but I'm not sure it's worthwhile given that you'd likely be spending 100% of the CPU time on the shaders anyway. I *am* curious if any of the weird LCD modes the CE's LCD supports would allow any portion of this to be offloaded.