Counterpoints to show that it is just a compromise: The complexity of unicode, collation, encoding, translation, language.
However you're 100% right!
Our application actually used text files on a network share with an indexer over the top as the database engine a long time ago. It worked really well and integrated with NT security and file locking for concurrency control, plus it was very easy to back up. A work of genius. However, NTFS doesn't scale well with lots of small files as it stores them in the MFT so it fell off a cliff eventually.
Counter-counterpoint: all of your counterpoints exist or have equivalents in every other form of communication. You swap unicode (which is nearly universally agreed on) for H.264/Theora/VP8 and AAC/Vorbis/Opus, and you still have to deal with collation and translation/language (which, without transcription to text first, is pretty hard).
Yeah we do that now. We store them in a big file in pages that contain rows and an externally visible network process allows us to manipulate the things.
However you're 100% right!
Our application actually used text files on a network share with an indexer over the top as the database engine a long time ago. It worked really well and integrated with NT security and file locking for concurrency control, plus it was very easy to back up. A work of genius. However, NTFS doesn't scale well with lots of small files as it stores them in the MFT so it fell off a cliff eventually.