I wonder what level of humidity that 2 dB per kilometer figure is for? Is that typical or worst case?

At 20 kilometers (about 12.4 miles), a distance this ISP is trying to serve some customers at, that 2 db becomes 40db. Overcoming an additional 40 db loss would require increasing the power by a factor of 10,000 times. They're using amplifiers and directional antennas but the fade margin is nowhere near 40 db. And they do have fading problems. Of course refraction/reflection from temperature/humidity inversion gradients and other propagation effects come into play too. It isn't just simple absorption.

Hoff, which graph are you looking at? I'm looking at page 20 of chapter 3 (http://descanso.jpl.nasa.gov/Propagation/1108/1108Chapter3.p...) - it does show a small peak but it looks to me as if it's at ~22 GHz, not 2.4. Note that it lies between "10" and "10²" on the X-axis. Also the magnitude of even that peak is only 2 x 10 to the minus 1 - i.e. 0.2 dB per km, and the nearby freqs are really not much different. The graph on page 21 agrees.

This article: http://www.martin.chaplin.btinternet.co.uk/microwave.html suggests that 2.4 GHz was chosen for microwave ovens as a NON-optimal absorption frequency, as this allows some of the energy to penetrate past the outermost layers of food!