I'm not a fan of "dumping" unstructured or BLOB data in-row in a RDMS either. The impression I get is that most users of NoSQL database solutions are basically just application developers who need some way to persist the session state of their web application, typically something like serializing the contents of a shopping cart as XML or JSON and then holding it temporarily until the user gets to the check out page. I'd actually prefer the website keep it's "stuff" in something like MongoDB. So long as the final purchase order gets properly stored in the RDBMS accountting system when the user clicks the 'Submit' button; that's all the business cares about.

I once worked for a company that provided a check imaging solution for banks. There was a transaction table, which contained a handful of columns like acct number, routing number, amount, etc. which were scanned from the physical check. The check image (a TIFF file) was stored within one of various folders in the file system. There was no "pointer" to the image, there was no path or URL inside the row. The actual path for containing the image was instead assumed based on client setup, and the image name was based on the row's unique transaction_id, which was simply an incremental integer or identity. So the storage of document images required no additional storage in the database.