We at the Hearst Museum use Delphi (http://code.google.com/p/delphi-museum-project/).  It doesn't link directly to our TMS back end, relying instead on static, mostly automated dumps from specific TMS fields.  The dumps are then processed using natural language processing and semantic algorithms, and then indexed against faceted ontologies to produce a faceted browser.  During the indexing, the available data is enlarged according to rules we've written into the ontologies (rules like "all xxxx are xxxx, except when they're xxxx and/or xxxx).  It's a powerful and exciting solution for data that are sparsely populated in certain parts of our collections.  

While Delphi doesn't interface with live TMS data, there are plans to integrate it with CollectionSpace (http://www.collectionspace.org/).

Delphi is open source software, but it's not yet at the point where a non-technical person could download it and run it without assistance (but we're heading in that direction).  Delphi makes use of some social tools like tagging and set building/sharing, and presents an accessible way to explore and discover our collections.  You can see the Hearst's implementation (still in development) of Delphi at http://pahma.berkeley.edu/delphi/ .  The part of our implementation that needs the most work is the custom-built ontologies; they require more work than I've been giving them.   In our implementation, we've also integrated our entire digitized card catalog to provide more information for the user for those objects where data entry hasn't caught up with our paper-based documentation.  We've gotten a lot of positive feedback from users who appreciate the friendly interface.  

Let me know if you have any questions about Delphi.  I'd love to see other museums taking advantage of this great system, and I'm more than happy to offer our work in ontology building as a starting point for anyone who's interested.

Michael