Environmental conditions could be playing a part here.
Flying over or near trees and other objects when it is
windy will subject the plane to eddy currents that
knock it off level. This plane will stay knocked off
until corrected.
Flying in calm over a paved parking lot will give
a similar effect when the plane hits the edge of the
thermal over the lot.
Since the plane falls off on one side then the other
it seems there is more than just the wind. The wind
would cause the first tilt, but after that it would not
immediately fall off the other side. I would first suspect
poor centering. Binding in the linkage would cause
poor centering to have a greater effect.
Bad geometry (using the wrong holes in the servo
arms) would compound the issue. If the linkage is
in the outermost hole of the servo arm, or the
innermost hole of the control horn, the
smallest increment of servo motion will have too
great an effect near the center of the servo throw.
Turning down the endpoints, and using expo instead
of using the correct holes only makes it worse because
the resolution of the signal and the servo resolution doesn't
change.
Finally slop in the linkage will allow the surfaces to
wiggle rather than stop in a specific position. This
is most noticeable in the center when there is less
pressure on the surface holding the linkage tight.
Again, this is magnified by bad geometry.
If correcting these issues doesn't fix the problem,
there could be an issue with the servo pot, or the
transmitter control pot. Both of these pots wear
more at the center that at the extremes. Servo
pots have the additional wear from vibration
(both engine/motor and wind turbulence through the
linkage), so it is the one to check first.
Finally the least likely, but most dramatic
cause of this behavior would be a loose
servo or servo tray. It wouldn't take much
movement of either of these to cause the
effect you are seeing, so it could be easy
to overlook.
Jenny