For the record, the author of those flowcharts is completely missing the point. You can just as easily call `git update-server-info` and copy the .git from your repo into a web-accessible directory, and just be done with it, the same as Darcs.
The reason people do it other ways (either via git-daemon, gitosis or ssh) is because Git uses a special transport protocol that's far more efficient than a dumb HTTP session. It's not that Git won't work over a dumb HTTP session, it'll just be a lot faster using it's own server/protocol.
All of the alternative tools - bzr, hg, darcs - have highly optimized protocols as well. This may be a documentation issue - I've never seen documentation for using a dumb server with git. Also, can clients submit changesets back via simple FTP? Darcs can, IIRC.
No FTP support, but that's insecure; what's wrong with using SSH? And there is documentation on the Git website (and in the source tarballs) on how to set up an HTTP-hosted repo, and even how to set up for a WEBDAV-enabled push repo, although I've never attempted the latter.
The reason people do it other ways (either via git-daemon, gitosis or ssh) is because Git uses a special transport protocol that's far more efficient than a dumb HTTP session. It's not that Git won't work over a dumb HTTP session, it'll just be a lot faster using it's own server/protocol.