Commented in Paddy's discussion of vanished reblog archives: it's not just "preserving the database" but tracking changes to post URLs & CSS.
...Rhizome has changed several times the way posts are organized and named --when you go from, say, a numbered post to having the word "reblog" in the url to removing the word "reblog" and assigning a content tag "reblog" to those old posts, links break and commands to "redirect" to the old links have to be written. You could go post by post and do this but usually programmers try to find a way to automate the process, so inevitably content gets lost. It's still in the database, but invisible to anyone using the site. Also, sites change their CSS design and information specific to the original post is lost that way, too. (E.g., removing a date stamp or a comment link by making it invisible in the CSS script.) Eventually you end up needing massive detective work to find posts (using Google's cache, other blogs, etc) and can never fully reconstruct a site as readers originally saw it.
Made belated response to commenter's "formalist net art about itself" in Paddy's "clubs to affinity" discussion.
...the surf club activities described in the Marcin Ramocki essay linked to above don't really constitute "formalist net art about itself," anymore than the science of linguistics is only concerned with the interplay of signs. There is always a connection to, and a concern about, the world these signs represent. When a web 2.0 artist talks, jokes, or makes art about art (or politiics or science) expressed in internet terms--i.e., reduced to jpegs and YouTubes--a comparison to the underlying "signifieds" of physical reality and history--their original meanings--is usually part of the equation. Not always, but with the better work.