Queries appear to be bypassing buffer cache

We currently have an issue on Oracle 11.1.0.7 on Windows 2008 64bit (ASM).

It seems as though every time I run a query, the buffer cache is being bypassed and the response is being read from disk each time.

Based on the Monitored SQL Executions, for example when I run the following query:

select action#, count (*) from sys.aud$ group by action# - the 1st time I run it takes approx. 25 seconds to run.

However, subsequent runs of this take exactly the same amount of time as well as the same amount of disk IO.

Looking at the query, there are very few buffer gets and a lot of disk reads.