Ways to mitigate this issue:
Upgrade to a newer version of Linux. (Issue is not present when using Red Hat Linux 6.4)
Force multi-block reads to use direct path rather than scattered read
Set “_serial_direct_read” to TRUE.
This forces all serial scans that exceed the small table threshold to use direct path read
This forces 11g style read semantics on the 10g databases
Change the Oracle init parameter “filesystemio_options”
The problem manifests only when this parameter is set to “SETALL”. Using “NONE” or “ASYNC” will mitigate the problem.
“ASYNC” is the preferred setting in the absence of “SETALL”