Even with chromaticities matching the bt.709 blue primary, significant disruptions in gradients are observed as the signal crosses the blue corner of the display gamut cube.
Significant disruptions and non-smooth behavior occur in that region at lower purity, and it gets even worse at higher purities between bt.709 blue and AP1 blue primary chromaticities.
These issues have been pointed out before and were not addressed much. See @TooDee 's post here, Nick’s visualization here, @jpzambrano9 's post here, and @meleshkevich 's post here
I think it’s safe to say that ACES 2.0 prioritizes invertibility and “reaching all corners” more than having nice image appearance. Those two design goals are strongly opposing I would say.
The nuke script that generated the above plots if anyone wants to try for themselves:
photographic_scene_001_bt.709_plot.nk (357.7 KB)