DXO has long been a joke in the photography world. People love simple numbers and summarized ratings though, it makes it so easy to say "mine is better than yours". Their absolute numbers reduced to a single score and weighting system are hilariously ambiguous, however there *can* be some value in their individual objective measurements if you look at them in isolation and take the time to understand exactly how their tests are performed before interpretation. This relates more to their camera sensor ratings, as their smartphone ratings are even more of a joke IMO (as you mostly point out above). There are numerous examples of their rating system failing, and going against literally every other objective result which has led to accusations of accepting money for scores.
Where the Pixel 2 is unmatched is mostly in categories that can benefit from Google's software and AI since the hardware is overall quite competitive these days:
HDR quality/ability
Low light noise (it stacks up to 9 images, and since noise occurs randomly, it can eliminate a lot of it)
Portrait mode (the dual pixel sensor works better than the dual lenses, and it deals with hair masking better. It also learns with AI and improves over time). The front camera does not have dual pixels and is 100% software so it's not quite as good.
Video stabilization (Google's EIS is ridiculously good, and now it has OIS on top of that) If you have a steady hand already, you can't tell the video isn't shot from a tripod.
There's also a subjective element because unless it's really bad, not everyone has the same opinion on an ideal HDR look, color, white balance, skintones, exposure/contrast, etc.
Google also has a "visual core" on the mobo that isn't enabled yet but will be for android 8.1. It is a custom chip made by Google specifically to improve photo processing (probably HDR). It will apparently make things even better but that's about all we know about it. Third party apps will also have access to it.
Before, the iPhone was at a major advantage because they used such small sensors, but now they are using more competitive hardware in that regard.