Nice to see you tried these insane settings
I wouldn't recommend it out of the box to anyone, only if you don't mind rendering a bit longer and for a mostly subtle difference.
The rendertime for this one isn't that bad actually.
I'm a bit surprised to see it didn't work out very well, to be honest.
My experience with this technique is mostly with the cubic b-spline which IMO benefits from this alternative approach.
In this particular case I'd say that the populations would benefit from slightly brighter textures (increase the diffuse colour value) and some more translucency.
No matter what sun angle and elevation, with GI 2/4/6 you should have been able to see detail in the shadows.
It's interesting anyway. For instance you could render it again with 8 atmo samples and see how that goes. Just out of curiosity.
You'll quickly see that you develop a feeling for the settings and that tweaking adaptive sampling can be a valuable option for optimizing objects/atmosphere rendering.
If used inappropriately it could also skyrocket rendertimes
Cheers,
Martin