as usually on slashdot: reading the original post does help: The NEPOMUK project did just that - id added the metadata database into the KDE system, on a level as deep as KIO or other backend services.

It is as if "on the operating system level", but technically not (because operating systems usually care about memory allocation, processes, threading, and other low-level business involving indivudal bits and not high-level filesystems or data)

Don't be afraid anymore - this was actually the plan of the semantic desktop all the time. For instance, a product of the project is http://aperture.sf.net/ which is built-in into http://eclipse.org/SMILA at the moment.

Also, Ansgar Bernardi (the project lead who was interviewed in the TR article) says:

well, I really hope this spinnoff company works out! ;-)

ha, good one.

There are some, but they are not very elegant:

http://dev.nepomuk.semanticdesktop.org/wiki/UsingNepomuk
http://dev.nepomuk.semanticdesktop.org/wiki/UsingDropBox
Or check out the KDE stuff:
http://nepomuk.kde.org/discover/user
also in cute little moving pictures:
http://www.youtube.com/watch?v=_8oavLQeAjM

speed. details. thats up to optimizers.
From what I know about all this, I would say that you get an open, w3c standardized destkop search engine. A slow fruit, but it is open and can grow, compare that to patented-getNervousBeforePublishingWinFS-microsoft.
probably the query thing is possible, but you have to learn the standardized SPARQL language first (oh joy) but wait and see:

someone who has programming skills and also wants to see the pictures of his wife and rating>4 will hopefully help us all by implementing the search UI :-)

/slashdot-humour=ON
... and of course there is a general interest in pictures of your wife rated 4+
/slashdot-humour=OFF