I'm testing for three weeks Nepomuk with Virtuoso using KDE 4.3.3 openSUSE packages, and finally thinks is working accurate. Or at least I don't detect bugs in my test .

I clean my strigi data following Trueg's blog indications, change to Virtuoso backend, and query performance was really good using KRunner or kio_nepomuk (alias nepomuksearch:/).

Query system lacks of power syntax, in part we can blame to SPARQL, but if you enable in KRunner Nepomuk Desktop Search Runner you can do simple queries easily and quickly.

The dark side was strigi because still is useless. For me, a system that don't create accuracy indexes or don't index anything has the same value: none. Current version can't do something as simple as extract Artist and Album from a mp3, so index your music is a lost of time and resources.

But, in the light side, I'm using Nepomuk to classify all my movies and series, tagged with language, country, actors, viewed or not, ratings and translations (I put translations as comments) and system works really good.

Some samples using KRunner with "Yuko Takeuchi" (the japanese actress in Flash Forward):

text:
hasTag:Takeuchi and hasTag:"no vista"
explanation:
I obtain all entries from "Takeuchi Yuko" that I don't see yet.

text:
takeuchi
explanation:
I obtain an entry with text "Open takeuchi yuko (tag)". If I click in it Dolphin opens showing me all directories with "takeuchi yuko" tag.

text:
hanaya
explanation:
I obtain an entry with text "Open 薔薇のない花屋" (hanaya is 花屋, flower shop in English)

And you can use these queries with "Folder view". In my case I have in my desktop a "Folder view" with query "nepomuksearch:/ hasTag:película and hasTag:"no vista") that shows me all the movies I don't see yet.

Seems to semantic desktop is finally ready to use with accurate results and in KDE 4.4 things probably would be better so, why not you give it a chance?. But, if you do, don't forget to disable strigi .