I read the MTL specification for materials and noticed that the specular exponent Ns must be a scalar value instead of a spectrum. Would a specular exponent with three color channels (i.e. RGB, XYZ, ...) make sense (i.e. could represent (the specular exponent is of course not physically based at all) something physically plausible)?
Using independent settings for each colour channel is like using settings dependent on wavelength when doing spectral rendering. So the question could be re-formulated as:
Is a material which changes its roughness with wavelengths physically plausible?
With some kind of nano-fabrication process you could build a material surface using some microfacets reflecting certain parts of spectrum and distributing each type of microfacets in a different way. So not only it is physically plausible, but it is also physically realisable. On the other hand, you normally don't encounter such behaviour in "common" materials.