If a query uses both indexed and non-indexed fields we should be able to execute it in two phases. A first less selective query on the index and then post process it by reapplying the whole filter again to results of the first phase.

This should be more efficient than completely ignoring the indexed fields and executing the non-indexed query on the whole data set.

A possible idea to derive the query to be executed in first phase is to apply the Boole-Shannon expansion theorem (http://en.wikipedia.org/wiki/Boole%27s_expansion_theorem).