I've already tested several versions (like you just with my.x+=3, fix c_move with time_step, fix c_move without time_step, vec_acc...)

The result is always the same - so it's not time_step related.