I experimented with FTS in MongoDB (I did what you did, except that I added stemming) and I was unsatisfied with the results. I think you should take a look at Solr. I was reluctant to add a piece of Java infrastructure to my tiny project (I am not a Java person) but that was the best decision ever in my case. Indexing is fast (compared to what we both did in MongoDB) and independent from MongoDB itself, and searches are much, much faster than doing it in Mongo. Plus you get all of the niceties of Solr: a ton of configurability, faceted searches, solid stemming in several languages and a decent user-facing query parser, all for free. Even if you don't need all of those right now, there is plenty of room to grow and most of it is a configuration file away from being available.
For our Jerome library project at the University of Lincoln (http://lncn.eu/fv4) we're using MongoDB to store catalogue, repository and journal items then we've implemented Sphinx (http://sphinxsearch.com/) on which we do the full text search, passing the results back to Mongo to retrieve full details.
We've found performance to be excellent, we can full text search over 350,000 records in hundredths of a second.
This looks more like key word searching, rather than full-text search. Were you able to search for letters in the middle of a word and match the word itself? I played around with something like this awhile back, using MongoDB, and concluded that using something like ElasticSearch (Lucene) worked much better.
Looks pretty close to the standard full-text Mongo example, but nice to see the issues they ran into putting it into real world practice. Looks like a surprisingly capable search for such a simple solution, minus a few things like stemming. Thanks!
