A small business has grown up using nothing but files. These files store large blocks of text which cannot be altered, just meta data pulled out (they're flat files, between 10 KB and 5 MB each, which is basically like dumping a large char[] arrays into files, and unfortunately they need to be kept in this format because the formats vary/incompatibilities/flexibility).
Anyway... Basically they have 30+ GB of these flat files. Literally millions of them. All in *.dat files. It works quite well like that. But the downsides as discussed in the link crop up and cause problems.
So we tried to put them into a few SQL databases as LONGBLOB, and after the first couple of gig the thing just became unmanageably slow, badly formed queries would cause it to lock up, and the thing took up way more space on the file-system than files did (and was WAY harder to manage with the performance issues discussed).
So I guess my point is, SQL databases are great for sorted data. They're freaking nightmares for CDATA/BLOBs. Just a complete waste of space. Almost no database is designed to handle that amount of data and continue to work.