I have a non-FFB stick, and I get the same problem. I suspect it's something along these lines:
1) I fly the helicopter onto a new attitude, 'fighting' the autopilot to get there.
2) I blip the trimmer.
3) The game logic, rather than holding the new yaw/pitch/bank attitude, holds my new yaw/pitch/bank control inputs.
4) My control inputs are ~20% greater than necessary, because I had to override the autopilot's old attitude hold inputs to get there.
5) The new inputs push the helicopter further in the direction I was maneuvering, because they're stronger than they should be, thanks to #4.
Holding the trimmer down disables the attitude hold functionality and gives you something like flight director mode, so you don't get the 'bump' when you fly onto a new attitude and hold it there—the controls are nearly where they should be for the new attitude, so the autopilot's authority is sufficient to hold it. There's another, unrelated jolt I get when I release the trimmer in the middle of dynamic maneuver. That one, I think, is something like this:
1) I'm pushing the nose forward, using more control input than necessary to hold the nose at my desired attitude.
2) I release the trimmer.
3) The helicopter jolts to the nose-down attitude where my control input would be stable, rather than attempting to hold the pitch angle attained when the trimmer was released.
In short, it seems to me like the trimmer is functioning exactly as a helicopter trimmer should, sans the autopilot (which is to say, I expect it to function like it seems to in Flight Director mode). The problem is, it seems to be functioning in the same way when the attitude hold autopilot is on (when Flight Director mode is disabled)—as a naive system to hold control inputs at their current positions, rather than as a true attitude hold system. I'm not sure if this is how it's supposed to work or not. If it is, I'll just have to stick with the decidedly non-Russian method of holding the trimmer down through maneuvers.