See if "include-tag" in the protocol can be changed to offload work to clients
@jacobvosmaer-gitlab has a patch to make the tag-following when fetch requests are served more optimized on the Git ML. I wrote a follow-up about how perhaps we can just not do this work at all on the server, needs some client changes: https://lore.kernel.org/git/87lfco801g.fsf@evledraar.gmail.com/