I have tried combining a best FWHM integration with a best SNR integration a few times in the past with a blend function favouring the FWHM integration for bright areas and the SNR integration for dim areas. Sounds very similar to what you've done, Lee.
I found that it works, but I haven't convinced myself that it gives a better result than applying a little careful noise reduction to the FWHM integration. If I was scrabbling really hard to find enough good quality subs I'd try it again, I guess.
I don't see why you'd get an actual reduction in FWHM. Sounds like you've just confused the measuring algorithm
Cheers,
Rick.