I've been reading up on UniWB and trying it out myself. My understanding is that the camera (and other RAW software?), in applying a white balance during the JPEG conversion, effectively overexposes the R and B channels (usually by a factor of around 2 to 2.5). The R and B channels may therefore appear clipped when in fact they are not saturated in the RAW data. Using UniWB white balance on the camera effectively sets all of these overexposure factors to 1, which means the histograms give a much more accurate representation of clipping of the RAW values.
However, one naturally never uses the UniWB white balance setting when post-processing. According to my understanding, applying a more sensible white balance in post may still clip the R and B channels in the result, just like in the camera. Thus, even though the R and B channels always are fully recoverable by drastically changing white balance or color settings (which may not be desirable), they may still appear clipped in the final result (appearing as so-called "nuclear colors").
My two questions:
If the above is true, wouldn't it be better to set a white balance in the camera as close as possible to what you will actually use in the final result, since this would yield a more accurate histogram for your final result? (I.e. colors that appear clipped on the camera, while possibly not truly saturated in the RAW data, will still appear clipped in the final result, so you might want to adjust your exposure.)
Applying a non-UniWB white balance in camera may show channels as clipped even though they are not in fact saturated (i.e. the histogram is overly conservative). Can this also work the other way -- i.e. can "normal" white balance settings in the camera effectively show channels as not clipped even when they are in fact clipped? (I don't think this is possible since by my understanding, this would require exposure factors below 1, which are never used, but it would be nice to have a confirmation of this.)