A Luminosity mask, aside from shielding the highlights, also introduces a weighting of G>R>B per pixel regarding the effectiveness of said curve
. Hence, bright yellow-green hues are receiving less HSB brightness compared to blue hues (of same initial HSB brightness). This counteracts an unproportional increase of Lab lightness with such yellow-green hues.
The issue of treating different hues the same is known with the HSB color model, however, it’s likewise valid for RGB curves. Whereas the perception of brightness is highly imbalanced around the color wheel – as is easily obvious from a 3D Lab gamut plot
where yellow-green hues peek along the L* axis. But, Lab lightness curves – even while controlling input/output in terms of L*, and even when being forced to preserve HSB hue & saturation (Simon’s way) – cannot really address this issue as long as the shape of a brightening curve is also subject of contrast requirements.
Do you now these “digital” images where yellow-green foliage somewhere in the background of a scene is by far the “hottest” part. Well, I think that’s the reason.
Thanks for the clarification.
I'm currently experimenting with yet another method to make tonal corrections without affecting hue/saturation by using the channel mixer in luminosity blend. So far, I'm rather encouraged. For a severely underexposed image, by dialing in combinations of channel values exceeding 100 in total, I was able to lighten it dramatically. While there are still some side effects like hue/saturation changes and loss of contrast, they seem to be far less than from using curves. They are easily fixed with painting in a layer mask.