Here it is mentioned that pixel intensity difference is used. Presumably this means we need to be representing color with HSV or a similar color space where brightness/intensity is its own dimension. Is bilateral filtering ever used in the RGB space where the distance metric is the L2 norm between the red, green, and blue values or pixels?

I can see L2 norm of the RGB vector being useful, though then you will have to define some threshold or normalization of the value. Which basically then you can then use a Gaussian function.

Bilateral combine the spatial distance and input image pixel intensity difference be able achieve the purpose of edge preservation and denoising.