That makes sense, in as much as something odd at the blue end in the reference spectra gives you trouble correcting the continuum. Do you know what the rough continuum shape should be for meteors? Any speculation on my part would more than likely be wrong.
I’m not sure if this will help, but sometimes I use BASS to process my spectra. This creates the response correction in a slightly different way. You still use your spectrum and a reference spectrum, but once you’ve done the division you use your mouse to select specific points on the divided spectrum. BASS then joins the dots to create a smooth response curve. I’m not sure if that would give you a way to create a response curve avoiding the dips or peaks.
Instrument and atmospheric correction is a tricky area. I’ve been struggling with it myself for the better part of this year, but I think I’m about there with my latest spectra.
Fascinating work you are doing. Like you say you can get lots of useful information even without the continuum correction.