It sounds as though you have SQL doing large number of "buffer gets" which could include consistent_gets, db block gets and disk reads(get blocks/buffers from disk)

I would look at the SQL first before muckin around with the memory. It is amazing how suddenly the hit ratios improve when SQL is optimised.

To get what SQL is doing "potentially" bad stuff try searching v$sqlarea for buffer_gets > x and then again where disk_reads > x where x is a integer eg 10000.

The other alternative is to trace active sessions and TKPROF to get SQL executed by that session and associated statistical goodies (consistent gets, disk reads elapsed time etc)

Also it pays to find out what the database is waiting on as well. Once you have solved contention and waiting then have a look at ratios.

SQL tuning, waits and then ratios is a better order to tune a sluggish system.

Have Fun