I get just about no spam at all (<10 per month, maybe).
Both result in other small mail providers getting blocked for arbitrary reasons with limited recourse.
You also either need to apply the greylisting to some larger IP range (rspamd e.g. apparently uses /19 by default for IPv4) or otherwise specially handle some of the bigger mail providers, because some of them rotate through their servers between retries, so you could be in for a quite a long wait if you do per-individual-IP greylisting.
The biggest culprit I noticed this with was Amazon SES – a former mail provider of mine used per-individual-IP, non-configurable greylisting, and any mail sent through Amazon (which isn't just Amazon itself – quite a few companies are using Amazon SES for transactional mail and suchlike) would consequently almost always arrive several hours late (however randomly long it would take Amazon to finally re-use an IP during a subsequent retry attempt).
Even more infuriating, my mail provider's support would then claim that it wasn't their fault and they didn't know anything about any supposed greylisting.
Mostly. But then you get the "click the link in the email within 10 minutes" problem. There's also a non-zero number of "our mail didn't get through first attempt, oh well, give up" people. From running GL on my servers over a couple of years, it mildly cut down spam (on top of blocklists and fail2ban) but I'm now wavering over whether it's worth the hassle.
That is the reason why I switched on some external block lists into the mix.
But much more importantly, that question is orthogonal to my argument. Using blocklists is a good way to cut down on spam, the fact that it might block some trivial percentage of people who for ideological reasons might or might not be on those lists isn’t the receivers problem.
You want this to not be the case? Contribute meaningfully to solving the problem in a more effective way. Don’t blame the people who just don’t want the spam.
I use blocklists in my self-hosted setup but only for the purpose of adding header fields that my Bayesian anti-spam filter can use to classify messages. I don’t reject anything out-right, aside from attempted spoofs of the domains my server is authoritative for. Everything is received— it just may end up in an “Unsure” folder if it seems too shady for the filter to put in my inbox.
