It uses an auto-generated binary index, which helps keep simple find operations fast. MongoDB-style find queries are naturally slower, but you can create indexes that are stored in-memory. Also, the authors have adapted it to suit mobile applications -- there are several special cases for Cordova, so it should work well for mobile-first applications. You can read more about performance in Indexing and Query Performance on the LokiJS wiki.
Other than indexing, there are other advanced features that projects like this usually don't support, including transactions and CouchDB-style views. Views are functions that can be stored with collection.storeView, then accessed with collection.view. There are examples for views, map reduce, and indexing on the LokiJS wiki.
LokiJS is tested with gordian, which is also written by Joe Minichino. There is a Gulp-based build script, and a demo application that's meant to run on the desktop.