Hi there,
you can use render tree like one in image. Just don't forget, once you got your image in render tree, it's in floating point, regardless of original format, so you can set colors to outside of 0-1 range. For example, you can create a 'fake HDRI' from 8-bit image. Also, Photoshop, due to it's nature, can afford some more sophisticated, but also expensive calculations - one example of that is a famous Photoshop's 'Soft Light' blending mode, which is completely different than one with same name in Softimage. On the other side, modern shaders can use additional calculations too, instead of simple additive or multiply.
All in all, you can't expect exactly the same result in both applications.
About setting expressions on animation divot for *color* input, this doesn't work in XSI for a long time. It's not a bug, it's just a "group" of three of four scalars, grouped together for easier keying, I suppose. For setting expression there, you can use three or four "Scalar_savestate" nodes, set expression on each one, then combine them with RGBA_combine node...
In your skin shading workflow, you used additive mode for blending layers. Mi_sss_skin is using "screen" by it's default. "Screen" usually means something like
1 - (( 1 - ColorA )*( 1 - ColorB )), and produces 'soft' look, it's hard to overburn image when using this mode. Result of blending won't go over 1 (if all inputs are inside 0-1), but it affects the blending at low values as well.
Screenshot of rendertree
Cheers
