I thought that one of the beauties of ICC profiles is that each profile characterizes one device. So for N devices we need just N profiles, all combinations of them are "calculated" at run time (notable exception being device link profiles). But using information in profile X to create a LUT of device Y seems to break this rule - would I then have to create as many display profiles as I have output devices? Maybe I got something fundamentally wrong about CM.
It's a fine theory, and works nicely for the device characterization, but gamut mapping isn't divisible this way, so it doesn't work for pre-computing gamut mappings. You can only do a really good gamut mapping if you know the source and the destination. It's something the ICC format hasn't really tackled properly, and one of the things that prompted the creation of WCS (Windows Color System). Graeme Gill.