[ODE] Cube and rectangular box reacting differently to torque

[Devrim Erdem]

Hello,

I am observing different behaviours when I try to apply a relative
torque to a cube and a rectangular box.

Both bodies are placed with a 45 degrees roll to right and applied a
relative torque of T=[0,0,1] (Z points up so the bodies are expected
to rotate around their own Z axis.)

The cube nicely rotates as expected but after some seconds the
rotation axis of the rectangular box is not its Z axis anymore but
changes continuously. For a moment it stabilizes at the inverse of the
original rotation direction and then destabilizes again. This loops
forever.

I wonder if this is a natural physical phenomena or a result of
accumulating numerical errors.

Could anybody shed some light on this issue ?

Thanks,

Devrim.