Very interesting stuff. I had a look at the "nexcepts" on your site. Yes, the Lingua derivational morphology modules (looks like Stem, Infinitive, Inflect) have provided some good results. It made me think about how I might go about doing something similar.
One thing that might make your searches better is some way to account for morphology that is not just stem + ending, like pronounce/pronunciation/pronouncement. Also, grouping (near-)synonyms like "brotherly" and "fraternal" may improve your results. Of course my examples are a bit textbookish, but I'm sure that you can refine things using your expert knowledge about what sort of information your clients might want to look up.