Currently the spectrum checking step loads in all the spectra, fits models, etc. - uses those results to determine which of the spectra are useful for the final fit, unloads all the spectra, and then loads the chosen spectra back in again.
I/O operations eat up time, and that would be particularly true for XRISM (for instance), so it would be neater to read them all into XSPEC once, and use the results of the spectrum checking to remove loaded spectra from XSPEC rather than to edit the list of acceptable spectrum paths and load that in again.