Putting the graphs next to eachother, the graphs looked similar, and after guesstimating the difference and applying that to the sensor as a correction in Domoticz, the graphs are overlapping sufficiently to explain the differences by the difference in location (one is in the center at ~1.7m height, the other on a shelf in a corner at ~2m height) and standard deviation... I'll keep it like this for a while and will reevaluate at a week's interval. FWIW I am actually doing the basic oversampling, while doing the timed single samples. If it remains an issue, I could modify the code to start the sensor with reduced oversampling/accuracy setting, but for now it looks good (enough)...

Could you perhaps try to introduce a few delays and see if that fixes the problem? Would be good if the sketch works without debug enabled.