Yes, I've been working with RDF(S) and OWL in the past 3 years. I find it extremely usefull for reasoning, and in general pretty cool to work with. Having all your data in a forward directed graph, including your Ontology (data structure) has some very interesting potentials. When I started with this project 3 years ago, I had to work with SPARQL 1.0 (the query language for triples/RDF), which was a bit troublesome sometimes. But now that SPARQL 1.1 has become a reccomendation, it is becomming well implemented, and working with it maked my queries quite elegant and fun again. Some people say the semantic web has long been dead since guys like google started their own semantic formats. Personally I think that's true at all. The vision/potential of the semantic web remains potent as ever. The tools are slowly but surely maturing. The field perhaps seems not that big, but it is surely active and growing. I think in time we'll see plenty of semantic web stuff entering the general field, and RDFa may be another helpfull bridge tool along the way.. Though in all honesty, I havn't worked with RDFa at all yet, so can't say much about that, but will surely have a look at this spec soon.

As an early adopter it was though (since 2008). However, if you are starting today it is not bad. And RDFa (lite) is a really easy way of getting in to it. Just make you site indexable using schema.org and RDFa for Google et al is a nice start.

I work for UniProt (a databases for biologists) and for us it is really a nice way of providing data to other users to integrate into their own systems. We also provide a sparql endpoint at beta.sparql.uniprot.org for other scientist to run analytical queries without having to deal with running a large datawarehouse of their own (beta.sparql.uniprot.org). i.e upload 5kb of SPARQL queries to us instead of downloading a 100Gb of data them to run SQL in their own systems.

These days there are a number of off the self systems that work well enough. Your main danger in the beginning is trying to deal with existing (large) public data before you are ready. The philosophical/reasoning side of the semantic web can also be very confusing but can be ignored in the beginning. Just use RDF as graph data format and SPARQL as a query language and you are good to go.

It's pretty cool. We use Apache Jena and some related Apache projects as our main SW stack, and it works pretty well. If you aren't real experienced with SPARQL, RDF, etc. and want to see some cool demos of what the Semantic Web enables, check this link for a taste:

http://wiki.dbpedia.org/OnlineAccess?v=6yg

The biggest fly in the ointment right now, IMO, is that working with larger datasets, with queries that require complex inferences via OWL, are crazy computationally expensive. And while there is research underway on the subject of parallelizing this stuff so it can run on clusters, I wouldn't say that that is a solved problem yet. So you can need some pretty honking fat software if you want to do really complex stuff.

I use it in a question answering system. DBPedia/Basekb/other random data sources backed by a search index and sparql allows you to do some cool things.