Chucka, there are have been some truly phenomenal hops done at microwave frequencies thanks to tropo thermal ducting:
July '94,July '91
Location: 33-43-45 N. ... 118-22-30 W.
Band: 2304, 3456, 5760 Mhz.
Distance: 2474.476 Mi. ... 3982.283 Km.
Talk about trapped in a layer!!!
You would need a very wide ack window to get any throughput at those distances, heh!
This was done via low power beacons continuously maintained on Mauna Loa in Hawaii w/ transmitter output of around 10w fed into some fairly high gain antennas, but the ducting is the real magic. Turns out there's a whole crowd of fanatics who love nothing more than hanging out and waiting for a duct to open. A little more more info here:
Veering slightly more on topic... As to avoiding nasty ducting effects, I've got a feeling based on incident angles to inversions/layers that very long hops might be more reliable if there's a fairly radical elevation difference between one end and the other, if you're in an area subject to frequent bouts of ridiculously calm weather. It would sort of figure that increasing the angle of incidence ought to help I suppose. Meanwhile, what exactly is the dependency on operating frequency is beyond my knowledge. We've got a 19k 5.8 link that's quite reliable but one end is at about 370m while the other is at sea level. We'll shortly be opening up another 5.8 link at 36km; this one has similar elevation differences but perhaps the reduced angle of entry to any potential thermal layer will make it unreliable. We very rarely have truly calm weather where this network is located so hopefully we will not find out.
All this stuff is undoubtedly researched down to a fare-thee-well thanks to the long history of telco microwave coms. Makes me feel like digging for it.