More interesting would be if a statistically valid sample of focus runs with one VCurve for all filters produces consistently poorer focus compared to using the VCurves generated from each filter. Measuring the focus is half the battle, as the HFD measurement itself is noisy. A better way would be to do a multi-star PSF fit then use the FWHM from that, using an image taken immediately after a focus run. If the differences are "small" then the seeing could dominate. But then it does so often... so what is the answer anyway :-)