Ok, I have had a chance to do some long range, NLOS testing and I have much better results to share with everyone about the SR9 vs XR9 questions.
First, this is my setup:
AP: 150 foot tower on a decent size hill.
SR9 is at 150 feet, 10' LRM400 into a 9dbi pac wireless yagi. Noise floor is -95 @ 907 MHz B mode.
XR9 is at 50 feet, 10' LRM240 into a 9dbi pac wireless yagi. Noise floor is -93 @ 907 MHz B mode.
So the first thing to notice is that despite being 100' lower than the SR9 and using lower grade LMR, the XR9 has 2 db worse noise floor.
The client is about 11 miles away. The path includes some lakes, and a mixture of leafy and pine trees. Testing was done on the rooftop (about 20 feet above ground level but still well below the tree heights) using two 12dbi Rootenna's with RB133C's at 12 volts.
The SR9 client had a signal of around -88 and a noise floor around -95. Throughput was an average of 200 kilobits/sec tcp. I turned on an old 900mhz cordless phone that I use for checking interference and the noise floor flew up to -75 and it somehow kept the link active (although throughput was pretty unusable.)
The XR9 client had a signal of around -84 (remember that this AP is 100' lower on the tower) and the noise floor was around -94. So it had better signal than the SR9 at the top of the tower... strange. Anyway, throughput was around 1400k/sec tcp. I misaligned the antenna a bit so I could get the exact same signal readings I had with the SR9. Once I had it bouncing between -88 and -89 signal, I tested throughput again and got around 600k/sec tcp. So the throughput was much better and would be the difference between being able to offer our standard 512k plan and saying "sorry, we can't reach you at this time"... (and yes, I know that it leaves no fade margin, but it is just for example). Anyway, I tested with the cordless phone again, and the noise floor went up 3 db to -91 and throughput hardly changed with the phone on. Amazing!
I did test the XR9 a little more in 5 mhz g-mode, and despite the signal going up about 6db, the throughput went down and the stability was lower too. I regret that I didn't do the noise floor test in this mode, but at this point, it was pretty dark and cold so I decided to call it a night.
So in the end, I am fairly satisfied with the results. If you're planning on swapping out your SR9's for XR9's, don't expect to double your coverage. It's not the "miracle card" that the marketing people at UBNT make it out to be, but it definitely gave me a solid improvement, especially for the customers in fringe areas. And regarding better signal levels, the XR9 seems to perform exactly as well as the SR9 when the signal was around -80 or better, so you won't see any better results there, but giving the fact that the interference didn't affect it as much, it could still make good links better!
If anyone has any questions, I'd be happy to try to answer them or try some additional testing.