LiveJournal has some fun with your feed, Sam.
Since this entry contains a linefeed but no HTML linebreaks or paragraphs, LiveJournal makes the assumption that yours is a dumb feed where the linebreaks are intended to be transformed into a BR tag.
I suppose what I'm suggesting is that if an entry isn't going to have any HTMLy linebreaks in it (which for LiveJournal means paragraphs or BR) it'd be best if you didn't put any non-visible linebreaks in the HTML either.
I'm sure LiveJournal isn't the only RSS client which is forced to make stupid guesses like this. Hopefully we won't have this problem with (almost-called-)Feedcast!
Martin: sigh. That's news to me. The problem is that every tool is making a different set of assumptions and guesses. Presumably that's because there is no way for a producer to convey their intent. That makes is nearly impossible to consistently produce a feed that can be properly interpreted by everybody.
The Atom 0.2 snapshot has a type attribute on content. There is talk of putting this attribute on summary, and perhaps even title and tagline.
LiveJournal's assumptions just boil down to a list of hacks because it seems that hardly anyone actually knows and/or cares enough to generate a proper feed. One of the silly things that has to be done is to override the default of UTF-8 with Latin-1 when invalid UTF-8 characters are found, because people just copy feeds with the charset set to UTF-8 but don't use software which makes this happen.
It also transparently maps to the Windows superset of the ISO Latin-1 charset if it encounters Windows-style curly-quotes and so forth despite the document declaring ISO-standard, which is also a common mistake.
The nastiest one is the newline one, though. I wonder how many feeds there are left which still generate HTML without linefeeds... I'm thinking maybe a better assumption these days could be "if there's any HTML at all don't add BR", still allowing for feeds which are plaintext.
I think all of the content elements in Atom/Necho/whatever needs some kind of type specifier. I know LiveJournal at least allows a small set of HTML tags in the 'subject' field (which is the 'title'), and will only be able to round-trip entries LJ to Atom to LJ if title can say it's escaped HTML. (I suppose I should put that on the Wiki somewhere, but I get lost on it these days and I'm sure someone wrote it somewhere already)
If the validator uses a compiant XML parser it probably already barfs on most of the things LiveJournal catches, since specifying UTF-8 and then including non-UTF8 characters should cause the XML parser to halt. In fact, I can see such an error in the FAQ.
The problem is that people don't validate their feeds. Lots of people forget their feeds even exist after they've hooked up their software to generate them. "Generate" usually means "spew out a bunch of XML and then iterate over the entries printing each verbatim", so it's no wonder that there are so many invalid feeds since lots of people don't even know what a charset is or understand the issues. LiveJournal's RSS feeds are often quirky because they get shortened and end up with half-tags and such. It's a pain to programmatically generate feeds on behalf of people who don't understand what's going on.
Your validator seems to cover everything I know about. I don't pretend to be an expert; The RSS support on LiveJournal is not by me, I've just seen a few error reports and seen your feed do some quirky things on LJ.