The high and low color is applied to areas the are above or below the range of the current z axis. The attached example has data for +/- 120, but the z axis is only +/- 100, thus you get areas shown in the hight or low color. Of course this only makes sense if the z axis is not set to autoscale. 😉
I think my confusion had to do with exactly what you stated - you must turn off z axis autoscale.
I can think of some good uses for this - a long time ago I wrote a VI to remove outlier points in 2d arrays/intensity graphs and color code them. This looks like it can be an easier and faster way to do the same thing.