This may due to the Nelder-Mead algorithm trimming to the wrong local minimum.
Is it possible to apply more lateral constraints when trimming?
e.g.
for straight and level flight - roll angle, yaw angle, roll rate and yaw rate all zero.
for steady turning flight - no lateral acceleration felt by crew - i.e sideslip bubble is zero. ( also met by required bank angle is a function of desired rate of turn and speed.)
for steady sideslip and asymmetric engine out, - yaw rate, roll rate both zero
When trimming 0.2 rad/s yawrate, does not actually turn when running simulate.