Re: WinFS = Document Management System not invented here…
What you're proposing would - if WinFS were to do what it is described as doing - mean that a "document" would end up as innumerable individual files on an NTFS file system, each one containing just one part (e.g. one span of text with the same formatting) of a document, and WinFS having an index of files that comprise "the whole document". If one were to offer a parallel interface to files - the normal one and the WinFS one - the normal one wouldn't show a "file" that is the document at all, just the innumerable individual files, one for each object in the document. I don't think that would have made any sense.
WinFS did in fact store some of its data as files on disks - but that was limited to unstructured data, and the unstructured part of of semi-structured data (such as images - unstructured bit stream plus structured metadata). But for structured data - they way it was intended to be used - that was all stored in tables in the database. What MS seems to have been aiming at was that all apps would expose the structure of their data to WinFS, and interact with it only through the medium of structured data storage interface on the database.
We've never really seen in general release anything that actually achieved this way of storing data. Gnome had a similar project for a while (GNOME Storage), but that seems defunct. Had WinFS or GNOME Storage been achieved, and all apps dealt with it purely through the medium of structured data, I suspect it could have been pretty performant. RDBMSs can be pretty performant when used well, and an awful lot of functionality (such as multi-user editing) could have become the default for all applications without the application developers having to implement it.
Bill Gates is said to consider that the failure to complete and deliver WinFS was a matter of regret. And I can understand why. The motivation - bringing universal meaning and standardised structure to all your data - was sound enough. Instead, what we generally have on all platforms (Windows, Mac, Linux, iOS, Android, etc) nowadays is myriad search bots, databases, files, other files that are also databases, all with their own code base, all tying up a lot of CPU time, all with their own vulnerabilities. On Windows alone I can count several - the Drive C: NTFS, the Windows Registry, the numerous databases that make up Active Directory, the file search / indexer's own database, the database storing local user account information. Heavens knows how many there are on a Mac. On your average Linux there's quite a few too; /etc/passwd, /etc/group, the ext filesystem, systemD's journal, /etc/shadow, /etc/hosts, etc.
Replacing the lot with just one database engine for the whole system (which is kinda what WinFS was heading for) makes quite a lot of sense. It'd probably be faster, and there's less room for security vulnerabilities across the system.