I'd love to see this if it's finished, it looks quite interesting.
A lot of good advice and feedback has been given here, but I want to weigh in on a few things.
As Martin said, the detail setting makes a big difference for relatively small (numerical) changes. When your detail scale is from 0-1, a change of even .1 (e.g. from 0.5 to 0.6) is large relative to the total range of the setting.
I hadn't noticed GI surface details being enabled in your original settings post or I definitely would have mentioned that. As again Martin said, it is seldom worth the render time hit and should generally be avoided. Basically I see it as a last resort when I am otherwise unable to get the detail I want with other combinations of GI settings. In other words, investigate all other options before using it, and definitely try renders - or at least crops - without it first, using settings that are otherwise "final". Only enable it if you feel something is "missing" in the GI (small-scale shadow detail), and even then you should do several small crops to make sure it makes a difference and a positive one.
Seeing the size of your object, I suspect that actually is also contributing to render time. If I understand correctly, you have a *several GB* OBJ file? That's pretty crazy, hehe. Do-able, but definitely taxing to the system.
The patch size no doubt also has an effect. I'd be curious just how much, and how much increasing it would affect your scene results visually. But with the high render time you're probably not interested in experimenting.
Finally, regarding raytrace atmosphere, I think it's worth saying that Martin is right in that it is not always beneficial, however I must disagree with his blanket statement that when using high(er) AA for vegetation, it automatically makes raytraced atmosphere undesirable. As I am just now finishing rendering of the Garden of Eternity animation, I had the need to do a lot of testing and render optimizations and my tests were unambiguous and detailed: raytrace atmosphere was faster for equivalent or better quality in all areas of this scene. This is using multiple volumetric cloud layers, one in particular that creates low-level mist, and is a source of a lot of noise.
Where I think Martin's approach may be making his own tests skew the other direction is the use of "Max Samples". I did try this with the Garden animation but, while it may be suitable for stills, I don't think it works as well for animation. Ultimately I used AA8 with 1/4 first samples and a 0.02 noise threshold. With this, using raytraced atmosphere was faster *and* less noisy, by far. I suspect if I was using Max Samples the results would be different however.
I also want to mention something I discovered while tuning the Garden scene render settings. *Increasing* the number of samples for a cloud layer actually *decreased* render time (significantly) when using raytraced atmosphere! This seemed counterintuitive to me at first, but I realized it's probably because of the adaptive AA settings. There is a relatively low noise threshold (0.02) and relatively high max samples. The AA sampling algorithm is therefore throwing more samples at the problem, up to its max level, in an attempt to reduce the noise coming from the cloud layer. Attacking the problem directly by increasing sampling of the cloud layer alone focuses the sampling where it matters, on the scene element that is creating the most noise. Thus you get reduced noise *and* reduced render time. In fact, at 32 samples, the quality is 2.8 for that layer! Yet it's faster than 0.5 quality. Something to keep in mind.
All that being said Martin is definitely right that RT atmosphere is not right for all situations. Hopefully we can come up with more concise guidelines on all this as it can be rather confusing and time-consuming to test lots of settings to get it right.
I do hope you'll post the results of this lengthy render!
- Oshyan