Shadowgin,
As I write here in a forum, each post is addressed to a wider audience with a mix level of experience, so the information here might be known to you, but I like to be complete, as far as possible.
OK, that was simple and as expected, but I had to see the set up. It can have many reasons, and guessing around is not really efficient. So, this was perfect, thanks for the scene file again.
The file for RGBA and for Multi-pass are at 8bit/channel. That might work to a certain degree, but not really when your scene is in Linear. Which results for even 8bit files to use a linear color profile, which is certainly far away from being optimal, talking about the Multi-pass file here.
The problem with 8bit/channel is simple, you get “bandings”, soon or later. To avoid that the files are dithered. If that works - fine, but for multi-pass, well, that is not really the way it works. Yes you can do that, but if all files would become dithers in a Multi-pass file, each differently of course, you would end up with a lot of noise. If all where the same, they would be visible artifacts. Not to mention that any dithering destroys any data channel. Multi-pass has many uses, but I guess if the final result, after finishing in post, will typically differ from the single image approach. Often strong changes are wanted even, and that needs a “fat negative” to say in film terms. You want information as differentiated as possible bore and after the changes. Low bit numbers limted this, hence why I write here so detailed about it. Check your histogram, you have only a small bandwidth here in use, and with the fine tonality in your image, in that small area the best differentiation is needed to gain good results.
Since a while I guess it is not even a good idea to use 16bit/channel for that anymore, even many people do that, and might get weird results perhaps. Again, it works to a certain degree, but the problems are just not easy to ignore. Speed, budget and other reasons are given typically, but on the end all that counts is the final product.
If you have on the other hand only a single pass, the Beauty pass as you mentioned it, this one gets after all the calculation in 32bit/channel/float(inside C4D) then the typical 8bit/channel dither treatment, if not switched off. When no post finishing is needed, that might be even OK. However, even small TV stations have switched already in the late ‘90s to 10bit as minimum, and yes, many people cheat here and it might even work.
The problem with compositing files, even a single (frame) multi-pass image in gamma space is kind of difficult.The advantage of linear are widely acknowledge, and since a decade+ even in Adobe’s main apps available (to a certain degree).
Most apps, like Photoshop or to my knowledge After effects, if not in 32bit/float mode, work in a “post-gamma” workflow. Light room BTW is pre-gamma, with an essential change how curves work and represents colors after a change, etc. (RED Cine X, has both BTW Pre and Post)
(I have done tests with your scene in all discussed/suggested formats. 8bit shows problems, anything else not.) So, do the same comparison with 32bit/channel float Open EXR and compare again, you will barely see a difference in tonal values. Barely? Well, since we have floating point many blending modes failed, anything like “1/a*1/b” kinda blend-modes (e.g., screen) and all mixtures thereof, as in Overlay. With the result that many companies, e.g., The Foundry moved to new formulas to get the wanted effect. The simplicity of the old days are gone, and with the new codes some derivations have been seen. Minor. Hence my carefully expressed comment.
I have said it often, and I do it again, the “.hdr” aka Radiance is NOT a 32bit/channel float format, far away from it. It is a 4*8bit -RGBE format, where the E stands for the multiplication of all three RGB at the same time. For colors the worse idea one can have to use it in production. So, OpenEXR. Which allows also to store in the best way from all formats to store data channels. Photoshop 32bit/channel works as weel, and perhaps a little bit better with the Adobe pipeline. I personally have a better experience with “.psb” files in 32bit/channel. Less prone to failure. Either way, even tiff works, they get rather large
So, anything 8bit, can be used, but it is not at all any kind of quality compared to what is used or requested. Yes, it is a delivery format for the web, but during production—it is simple not anymore.
I hope my detailed text will help to avoid the problems you mentioned. The question why is it offered anyway, because even today where we all move toward a “REC 2020 UHD HDR TV” signal, we work with proxies. Or just for motion and animation tests, etc.
My best wishes