Some tests later.
What I did:
I made an effect that sets and resets the Channel of one of the LED in the B-Station.
I tried:
Jump to 100%, hold 0,2 s, jump to 0%, hold 0,2 s ...
rise to 100% in 0,1 s, hold 0,1 s, fall to 0% in 0,1s, hold 0,1 s....
rise to 100% in 0,2 s, no hold, fall to 0% in 0,2 s, no hold...
All the three worked and showed up as well in the LED as in the Artnet-Viewer connected to the same switch.
And I even recognized if there was a ramp or a jump in the value.
Another test was done with jumps to 100 and 0 with 0,1 s hold and ramps of 0,1 s up and down without hold. Even this signal with same frequency was transmitted and I could recognize by eye the different behaviour of the LED and the display of the analyzing tool depending whether I sent a ramp or not.
So I assume the artnet refresh rate is much better than 10 frames per second. And I am really sure much much better than the 0,1 frames per second you assumed makes B-Station think Artnet Signal is lost.
One thing changed: During that flickering, there was no dropout. LED followed the signal changes permanently.
Nevertheless, B-Station misses recognizing artnet signal when receiving steady signal other than 0%. I think I recognized some kind of rhythm: 0,5 s LED is brigt, 1,0...1,5 s it is dark. This is synchronous with the "monitor" display of the Web Frontend of the B-Station.
Concerning the switch: All possible functions (as QoS, Loop detection, broadcast blocking) are deactivated.
PC with analyzing software and B-Station now use both PoE-capable ports on the same switch and I even crossed them. No change in behaviour.