the Cisco APs have a short distance between the two BNC connectors for 802.11a and a longer distance between the BNC connectors for the b/g antennas.
I once read that these distances are chosen according to natural constants. Does anyone know the exact story of this?
It's predictable pattern.
Assuming the usual "rubber duck" antennas:
If they were a half wavelength apart, they would nullify each other(180 degrees out of phase)
If they're a 1/4 wavelength apart, the pattern becomes enlongated between the antennas (think of the antennas as "goal posts" and the generated signal pattern is stretched int eh direction of the ball going through the goal posts).
At one wavelength apart, the general pattern is still generally circular / omni, and there is some wavelength magic math that suports better propagation.
So, with a cable-connected external antenna, not a big deal ... they could be any distance apart; with a direct-mounted rubber duckie or whip (or dipole, or , or, or) you get some wave magic (or not, depending on the antennas) by starting at one wavelength.
Note: if you want to PREVENT "coupling" when you mount an antenna to a conductive surface, you should put it (IIRC) at least TWO wavelengths away.