Overall, query response time dramatically decreases with the use of sets, therefore it is preferred that user selections (or the cube itself) prevents the potential of choosing multiple members within the same dimension.
Queries that only included 1 member in the slicer but are still represented as a set are not preferred. Change the query so that when 1 member is in the slicer, that choice is represented directly within the where clause (foregoing the set creation/utilization step altogether), this alone will dramatically improve query time.
Also, ensure that the use of sets is absolutely necessary to achieve the query results that you desire, otherwise it's slow going.