My name is Conal Tuohy I live in Wellington, New Zealand. I've been using Cocoon for a few years, and I like it a lot.
I work at the New Zealand Electronic Text Centre which has a Cocoon-based website, and also offers consultancy and text-processing services to external clients, generally using Cocoon as a platform.
The NZETC website uses ApacheModProxy so as to serve most of the site from static files, and other parts dynamically from Cocoon.
The site also makes heavy use of MetaStylesheets for HTML templates, and for Schematron validation.
The site uses Batik for scaling images and making thumbnails.
I wrote the documentation for the LuceneIndexTransformer and also changed the transformer a bit so it can update an index incrementally. Now we use this component for indexing the NZETC site. It's a far better way to index a website than with a crawler, IMHO.
I also wrote the JarProtocolExample HOWTO which describes how to read data from inside a JAR, ZIP or Open Office file, though personally I haven't used this at all
The site is also based on a Topic Map. We use XSLT to harvest metadata (in XTM form) from our source documents (TEI-encoded books, and a MADS authority file). We use TM4J to merge the topic maps together into a mega topic map (about 180Mb of XTM). Finally we use JXTemplateGenerator to query the map (using jx expressions based on TM4J's API) and present the semantic network of the topic map as HTML links etc.
The idea is that the Topic Map provides an architectural layer to separate the raw resources from the page layout and navigation, and that this facilitates a better division of labour between the two distinct concerns, and also allows for a very high degree of inter-linking between related resources on the site.