See: "Also, when we block an IP range for abuse, innocent third parties can be affected."

Although they refer to IP ranges, the same principle applies on a smaller scale to a single IP address: (1) dynamic IP addresses get reallocated, and (2) entire buildings (universities, libraries, hotels, etc.) might share a single IP address.

Aside from accidentally affecting innocent users, you also open up the possibility of a DOS attack: the attacker just has to abuse the service from an IP address that he wants to deny access to.

More sophisticated client identification can be used to avoid that edge case, e.g. TLS fingerprints. They can be spoofed as well, but if the client is going through that much trouble, then they should be treated as hostile. In reality it's more likely that someone is doing this without realizing the impact they're having.