is it then correct that problems start when the calculation cycle is (a lot) outside the rendering refresh rate ?
Yes, that's correct. But not only that. The Windows timer is a very crude instrument with precision granularity as low as 15~16 milliseconds. Which means that
timeOut incrementation of 1 to 16 above ~16 milliseconds will immediately reset it to 32 milliseconds, 32 + 1 to 16 milliseconds, to 48 milliseconds, etc. etc. etc.
The Windows timer should be used only in simplest OpenGL applications with a
fixed FPS rate that is a multiple of 16 milliseconds. On the other hand, the timer is a very easy means to minimize CPU load in immediate-mode OpenGL applications.