Isn't there any possibility to tweak the contrast or the tone curve of the individual color channels in addition to balancing the levels?
We have to distinguish between RAW data and data that has been mapped to a target color space (e.g. Adobe RGB or sRGB).
When you mix channels or apply curves to them in the RAW data, you get the degrees of freedom that are typically used for color profiling.
This is significantly more than just white balancing. White balancing tries to equalize the channels in a single target.
In color profiling you have a bunch of target, e.g. a color checker chart and you then solve an optimization problem where you try to minimize the deviations of each patch from the target value. Then there are different ways of doing it depending on the software, e.g. DNG profiles in ACR. They are actually computing something that does not interfere with white balance. So you can apply the profile but have to set the white balance regardless.
When we talk about the data of the exported RAW conversion that is in sRGB or similar, even applying a white balance involves nonlinear transformations.
For example, when you use the grey eyedropper in the curves adjustment in Photoshop, you will observe that a different curve is applied to each channel.
This is one reason why it is much better to white balance the RAW file instead of the exported RAW conversion; for RAW files, the white balance is a mathematically simpler operation that does not stretch or squeeze anything.
By the way, I'm also interested in the comb-like pattern on the large and wide peaks in the histograms of the blue and the red channels. I wonder if they would suggest the posterization or lack of gradation due to the fewer pixels for these colors?
Do you mean this stuff?
This is simply an artifact of the histogram building process. Let's zoom into it a little bit and show only RAW levels 64 to 1024 (i.e. -8 stops to -4 stops relative to saturation)
The histogram is built on a log scale. Since the levels are integers (1 through 2^14), the bins in the log domain (1/96 EV in this case), don't map to even levels. For instance, let's consider the first bin from the left that is standing out (just right of -7EV). That bin runs from 148.957 to 150.036. So the pixel counts at level 149 and 150 are counted as belonging to that bin. The bin just left of that runs from 147.885 to 148.957; so it counts only pixels that had a level of 148. So even if there are the same number of pixels each for levels 148, 149, and 150, the right one of the two bins would have double the number of counts. You can check for yourself that the count is exactly double of what it should be.
If you display the same range on a linear scale, the problem is not present:
Here, the bin size is 1, so all the little jaggies in the histogram are 'real'.
The comb-like patterns are purely a matter of histogram display, and have nothing to do with either the scene or the camera sensor.
