I am dealing with these types of sensors on a professional level so, I thought I could add to your experiments.
As mentioned above these sensors are very sensitive to organic compounds, which get emitted by silicon bases materials.
This can go as far as tainting the sensors completely and rendering them useless. I remember a oversized project to qualify new glues for the whole production, after having to scrap 5k+ sensors because of tainting. They said the sensors where “poisoned” 
A standard approach is to get rid of some residual compounds is to heat the sensor to a higher temperature to get rid of these compounds, then lowering the temp. to get a reading.
This is all not low power application but I guess BOSCH is going this approach. The reason why you get different resistive values after you experimented with preheat phase times, seems to me as a typical finding after a longer heating phases, which “cleaned” the sensor.
The reason why you get higher values with longer preheat times is weird though. Maybe they use the known thermal mass of the sensor to estimate the die temperature from the heating time with known power. But this would be a very crude approach, but would explain your results. The longer the preheat phase, the hotter the senor, changing your results.
