I always thought that Google has a huge competitive advantage here, because most people browse the web being logged into their Gmail accounts, and, therefore, as with Google Analytics and Google Adsense, Google knows that it's you who is viewing that page. It can then present extremely time-consuming CAPTCHAs to anonymous visitors, most of whom are likely to be bots or the spammers themselves.
You want Google to not know about you. You want to be a stranger to them. And you are complaining that they don't trust stranger, which you want to be, as much as someone they know?
This has already happened with tor and Cloudflare, but at least that changed for the better recently (see https://www.zdnet.com/article/cloudflare-ends-captcha-challe...). In that case it was just one CDN using captchas to discriminate against a group of users, so that one change by the CDN could fix the issue. If too many random sites are independently blocking or slowing down anyone not logged into Google, then that'll turn the web into Google's web.
I'm so bloody sick of helping Google train their self driving cars. I swear I'm going to have PTSD about stop signs and store fronts for the rest of my life.
> I've started deliberately getting them wrong - probably won't make a difference, but makes me feel a little better about it
I don't even have to try anymore to get them wrong on a regular basis. Now, I think it's now more like training Google users to make the same recognition errors as its self-driving cars than training the cars to do a better job.
I can only fathom these shops, both management and the webdevs, have no idea how unprofessional their site looks to anyone that isn't using a vanilla ISP connection. And my experience is coming from using a single longstanding VPS address, not even a shared VPN.
A sensible scheme would allow a certain rate of login attempts per any IP before hassling a user, but Google is obviously more interested in getting their training data than making sure you don't lose customers!