After some discussion, I believe the luminance weights for ACEScg in Natron are correct.
For posterity:
To calculate luminance weights for some RGB colorspace, one would calculate an RGB to XYZ matrix using XYZ Scaling as the chromatic adaptation method, and then take the middle row of that conversion matrix.
For example for converting ACEScg to XYZ E w/ XYZ Scaling CAT, the matrix would be
0.695383 0.140665 0.163951
0.272229 0.674082 0.0536895
-0.00552588 0.00402521 1.0015
(calculated using my GamutConvert tool)
This yields a luminance weighting of 0.272229 0.674082 0.0536895
(rounded to 6 digits of precision).
I’ve pushed a fix to my Saturation tool to reflect this.
As I was typing, Thomas has confirmed my hypothesis. Further trust this is the right answer!