After all, isn't green always a coefficient of 1? Why should the pixel values increase?
You can see blowouts in the green channel with UniWB that you didn't see a regular white balance because "green" doesn't just live in the green photosites and some green was being calculated from red/blue photosites.
There's overlap in the spectral response of photosites, so even red and blue channel photosites have some green component to them, and they are the values that are being "swung" by the white balance settings.
Let's say that there was a fair amount of green in your red channel and the coefficient was .8 for a regular white balance. That would understate the green a little. Now make that red coefficient 1 and the green also goes up in value slightly.
It's actually far trickier than that simplified description. Remember, we're dealing with data after the demosaic has occurred, so what the demosaic routine is doing with neighboring photosites is important to the end result.
Bottom line, even UniWB isn't a perfect representation of the individual photosite raw values—it's calculating off the embedded JPEG after all—but it is far closer to the truth of the raw data than any other white balance setting. For those of us paying attention to every last detail in our data capture, what I call "capture optimal data" instead of "photograph", it's the best we have.
