Whatever the base file formats are, Project Gutenberg, like most archives, needs to pick one or a small set of them so that the people who use Project Gutenberg can know what they need to read the files.
Again, this is confusing input file formats with output file formats. PG could choose to allow HTML as an acceptable input file format because PG can easily write a tool to convert HTML to their choice of PG TXT file format, including standardizing on such issues as whether italics ought to be rendered in PG TXT files as *star* or +plus+ or _underscore_ or SHOUT or better yet maybe PG could allow these kinds of choices to be made by an output filter so that text readers for the blind could have something more compatible with their prosodic emphasis machines, or better yet maybe the output filters could actually implement some of the "proper" prosodic emphasis markings for the more popular blind reader machines in order to maximize their capabilities. In my experience what happens is just the opposite of what you might expect -- rather the first time user of PG picks up a PG TXT file because they think that represents the "lowest common denominator" for their machine and so they think "it must surely work" and what they find instead is that what gets displayed on their machine is a total hash of line breaks in non-sensible locations, and random garbage marks, and then they conclude PG is archaic brain dead stuff by people who are clueless and they give up and go away. Or alternatively they post stupid stuff on public forums like "gee I like all these free books from PG and I read them all the time even though they have these random line-breaks stuck in all over the place" -- which in turn makes the efforts of the PG volunteers look like clueless idiots. There are other sites which take PG texts and do intelligent things like "tell me what kind of machine you are reading on and I will suggest which of the many file formats will probably display to your liking on your machine" which I think in practice tends to result in happier customers. Right now PG is still basically assuming that the average PG "customer" is a die-hard hacker running some flavor of a *nix machine in a college environment. Which is probably [somewhat] true of the people submitting books, but not at all true of the people who would just like to read them.