It is resource inefficient, which is why PiHole supports it, but does not allow you to subscribe to list containing regex, as that would quickly make it unusable. I'm not sure how Dnsmasqs would be accomplishing this feature without some sort of pattern matching logic - which would have the same inefficiencies. Basically O(N) where N is the number of domains/patterns that should be blocked. There could certainly be a cache to keep track of matched/unmatched queries, but I would imagine with the modern web the average case would still be very close to O(N). Just speculation.

Dnsmasq 'address=' function is just a substr() call, which is as fast as 'normal' hostname blocking (host == "adhost.com"). No regex magic is required there.

You are not able to block something like 'ads.%.adhost.com', but only (prepending wildcard) '%.ads.adhost.com', which in practice will cover almost all scenarios where random subdomains are used by adhosts.

If the regex rules do not require backreferences, it should be possible to implement them in such a way that they run in constant time with respect to the number of rules.

> regex is _extremely_ resource inefficient

No, it's not. Is there something wrong with their implementation?