The buffer cache advisory is implemented as a initialization parameter (db_cache_advice) & dynamic performance view (v$db_cache_advice).
After turning on the db_cache_advice and running the database under normal conditions for a while, query the v$db_cache_advice view (see the supplied script db_cache_advice). This will produce 20 rows of output which will provide estimated hit counts if you increased or decreased the buffer cache:
SIZEMB SIZE_FACTOR #BUFFERS EST_P_READ_FACTOR EST_P_READS
------ ----------- -------- ----------------- -----------
4 .1667 500 4.72 4,463
8 .3333 1,000 1.08 1,019
12 .5 1,500 1.00 945
16 .6667 2,000 1.00 945
20 .8333 2,500 1.00 945
24 1 3,000 1.00 945
28 1.1667 3,500 1.00 945
32 1.3333 4,000 1.00 945
36 1.5 4,500 1.00 945
We can see from this partial output that on my test database, the current cache size is 24M (size factor = 1), and we can reduce it to 8M without much negative impact (only an increase read factor of 1.08).