We use a theoretical spectral library of stars for flux calibration.  All the spectra have a linear wavelength grid and the wavelength can be described using CDELT, CRPIX, CRVAL header keywords, i.e., there is no need to store the numbers.  The total file volume of the library is still not clear (there is a lot of room for reduction), but in order to further reduce the volume, I would propose to use these header keywords to describe the wavelength of the model spectra and store only fluxes in the fits body.

I do not know who should be assigned to this ticket, but I assign RHL for now.