Can you flesh that out more? In the case of AI scrapers it seems especially clear: the model companies just want tokens, and are paying a (one-time) cost of C for N tokens.
Again, with Hashcash, this isn't how it works: most outbound spam messages are worthless. The point of the system is to exploit the negative exponent on the attacker's value function.
Perhaps I caused confusion by writing "If botPain > botPainThreshold and humanPain < humanPainThreshold then Anubis is working as intended", as I'm not actually disputing that Anubis is currently ineffective against bots. (The article makes that point and I agree with it.) I'm arguing against what I take to be your stronger claim, namely that no "Anubis-like" countermeasure (meaning no countermeasure that charges each request the same amount of CPU in expectation) can work.
I claim that the cost for the two classes of user are meaningfully different: bots care exclusively about the total CPU usage, while humans care about some subjective combination of average and worst-case elapsed times on page loads. Because the sheer number of requests done by bots is so much higher, there's an opportunity to hurt them disproportionately according to their cost model by tweaking Anubis to increase the frequency of checks but decrease each check's elapsed time below the threshold of human annoyance.
The scraper breaking every time a new version of Anubis is deployed, until new anti-Anubis features are implemented, is the point; if the scrapers were well-engineered by a team that cared about the individual sites they're scraping, they probably wouldn't be so pathological towards forges.
The human-labor cost of working around Anubis is unlikely to be paid unless it affects enough data to be worth dedicating time to, and the data they're trying to scrape can typically be obtained "respectfully" in those cases -- instead of hitting the git blame route on every file of every commit of every repo, just clone the repos and run it locally, etc.