I use them for natural language processing. There are cases where the performance increase is enormous (as in I wouldn't be able to afford the computation without them) and the downside of false positives is or can be made irrelevant.
https://github.com/RaRe-Technologies/bounter#example-on-the-...
Reduces RAM requirements some 31x times when automatically detecting common multiword expressions (e.g. "New York", "network license" or "Apache Hadoop"), which is significant.