It could only be used as a tie-breaker for search results with the same level of confidence, anyway.

It would be ridiculous to down-rank the exact thing the user is searching for just because the user would have to wait 800ms longer for that information. Or up-rank something the user isn't looking for just because it loads faster.

The best Google can do is bluff about how much perf matters.

The efficacy of the incentive is linked directly to the strength of its effect. If optimizing the hell out of your company's site only matters in extreme cases where it's a tiebreaker among hundreds of other signals, the people who want the things that make pages slow will win. They will be able to point to more tangible and measurable benefits, and the effect of the tiebreaker will be lost in statistical noise.

It may just be unfounded cynicism on my part, but this does not sound like a better web experience. It sounds like the web circa 2009-2015. It sounds, to me, exactly like all the things we'd like to get away from with something less intrusive than AMP.