The issue was brought once more in Dan Margulis "Color Theory" list. I think it might be of interest here too. Magenta is lack of green. The reason behind magenta tint is multiplication of raw data on white balance coefficients in red and blue, as well as data truncation due to bit limit of the result. Here is how it may happen. Let's assume the highlight recorded by the camera is (R,G,B) = (2100, 4095, 3011) in 12-bit (max = 4095). White balance coefficients recorded for particular light are (Cr, Cg, Cb) = (1.95, 1, 1.36). Performing multiplication you will have perfect white - (4095, 4095, 4095). Now, let's see what is going on when green channel is overexposed. (R, G, B) now can reach for example (2835, 4095, 4064). Red channel and blue channel had some room to grow, while green was already saturated. Applying same coefficients we have (5528, 4095, 5527). As you see, red and blue channels are balanced, while green is lacking. Hence magenta tint. To avoid this tint we can truncate values to 4095, or to use different "tricks" to restore green. Another issue with false highlight coloration arises due to two-stage approach to applying white balance in some raw converters. If highlights are already truncated at first stage in one or two channels when camera white balance is applied, on the second stage, when we use resulting data for manual white balance tweaks, highlight false colouration may be grossly amplified.