No, it would not. Spammers are still paying for the devices - most botnets are built and used (to spam) by different actors. Systems that require more computational power to spam take up more of their resources, making spamming significantly less profitable.
Depends on how much do you want real users to spend per captcha in electricity cost? If say it is $1/captcha it would be untenable for real users. If it is 0.01 cent, it wouldn't hurt spammer's margins.
Yes, $1/captcha is clearly infeasible - it'd be far too slow, first of all.
I don't have any knowledge of what spammers' financials are like, but it's possible that even 0.01c/captcha would still be impactful if the click rate is low enough.
Probably the best way to start tuning the PoW difficulty is just by starting out with what users are willing to tolerate - e.g. 3s solve time on the median mobile device. The gap between mobile and desktop devices has significantly lessened over the past decade, so desktop-grade equipment won't have that much of an advantage - say 1s per captcha, which is a lot for a spammer who would otherwise be able to send out dozens of spam per second.
It's not about making spam impossible, but about making it unprofitable enough that the criminals go elsewhere. Economic warfare.
According to google search, captcha solving companies charges 0.3c per captcha[1], which basically translates to half an hour of PoW for digitalocean instance. So if the PoW is less than half hour, spammers would need to pay less with PoW.