Yes, that's part of my concern - that the data might be installed "internally" by the software in some highly-encoded form such that it's difficult to recover from a crash (and these sorts of issues tend to get worse as time goes on).
In the event of some major catastrophe I don't think I'd mind too much if the audit trail of who created/edited what and when was lost - but it'd be nice if the raw article text (and images) could be migrated to a different system relatively easily if it were ever required. I expect some flavours of wiki software handle that aspect better than others.
Tim mentioned "RDBMS and a bunch of files" though, so perhaps it's nice enough to just store the "audit" part in the DB (and how articles link to each other) with the actual data kept seperately as plain-text.
cheers
Jules