I found why MT reads random voltage...
If you look at the schematic you can see that booth of the nernst channel is connected to the OC2 (atmel pin 17) thru a resistor and a cap.
On that OC2 pin, there is a ~3,5kHz squre wave.
Now I wonder why it needs that squre wave? Is that needed for WBO2?
Would it be good, if I put a pulldown in the O2 signal path, then the opamp as a buffer, after that an RC filter?
This would be something similar like in V2.2 (In V2.2: O2 input pin -> pulldown resistor -> RC filter -> ADC input.)
The input pin is EC18-1, that is sure.
One more thing: maybe I got this "problem" only because the (heated)sensor can not warm up to the operating temperature on test bench? Maybe after it reached that temperature this squre wave won't bother anything...