By the output of
vmstat -v's
numclient it seems there would be enough cached files that could be discarded to get free ram.
The AIO requests do not hit the maxreqs so that's good too.
By the mount options it looks you are not using DIO (Direct IO) nor CIO (Concurrent IO) which is often used to get some performance increase.
We implemented CIO once and it was a huge increase on performance for the Ora DB as the filesystem buffers are not used anymore and Oracle handles it itself.
Implementing it I can not say if it changes to the many blocked threads in the RunQueue.
Though if there would be a need for better performance, this should be a way to go.
For implementing it see the appropriate documentations as it may only be used for some types of Oracle files ie. filesystems.
Another much way, which is a lot more work than just adding mount options and the setting agent.kgb mentioned is using the Oracle ASM where you supply raw devices and Oracle will handle them accordingly. Though I have no experience with it and I am not aware what is the current "state-of-the-art".
Here is a document about implementing ASM:
>>>
IBM ASM Doc
For the short glimpse and not knowing if there is some workload peaks, the box seems to could have a bit less CPU and RAM.