The almost universal opinion is that HS-65s are the only way to go for cyclic. Now, the HS-65 is a fine servo (I've got them on my Trex) but they're fairly expensive and use a lot of current, forcing the use of a pretty hefty BEC if you want to be conservative. So I found myself wondering if all that torque is really needed.
There's a nice servo torque calculator at
http://www.csd.net/~cgadd/eflight/calcs_servo.htm but it's only for fixed-wing. But just to get an idea I assumed that the rotor blades were just a big control surface travelling at the tip speed (which is not right since the root of the blade is going much more slowly than the tip, but it gives us some margin.) For typical Trex blades, 2*325 mm long with a chord of 32 mm, 2400 RPM so about 320 kph, and 10 degrees full pitch at 30 degrees servo deflection, that gives about 12 in-oz of torque, which presumably would be split across all three servos.
Compare that to the HS-65 max torque of 25 in-oz (31 in-oz at 6V) and it seems that the 65s are overkill for the actual loads.
Now this ignores friction in the system and any effects of the lever arms in the linkages (which probably only help) as well as the fairly small additional load of the flybar paddles, but it seems that a typical mini could get by with no more than maybe 10 in-oz of torque on its cyclic servos. This may explain why people were happy with HS-56s, and suggests that even people using HS-55s or other small servos are not crazy (though the HS-55's speed is poor, and that's another factor.)
If you run at 3000 RPM and +12 pitch, then the torque load goes up to 27 in-oz, so people flying in that range may really need HS-65s.
What I'm not sure of is what the transient load of a fast cyclic input is. That's probably when you want your servo to not stall.