Thanks a lot for your answer.
Using fix nvt, the oscillations vanish. The xz tilt increase is constant over several flips and the numerical values of the temperature etc. coincide perfectly with the simulations with "flip no".
"I'm thinking the noise in xz is due to the fact that NPT is continuously adjusting the x and z box lengths (independently) hence the instantaneous xz tilt is non-monotonic. "
Do you have any idea why the noise only appears as soon as the cell is flipped at least once? As long I suppress flips, there is no noise.
What makes me especially uncomfortable is that also the numerical values of the temperature and the pressure deviate as soon as a flip occurred (although I am not sure whether the deviations are significant since long simulations with no flips end up very fast in an extreme tilt).
"What you really care about are the box boundaries for the deforming box. Are those more smooth?"
I am not certain, whether I get you point here. What do you mean by box boundaries? xlo, xhi, lx, ylo, yhi, ly etc? From my understanding the xz tilt should be independent of these values as I use fix npt only in xx, yy and zz and set "scalexz no scaleyz no scalexy no".
Best wishes and thanks once again