There is no such thing as "optimal" once and for all. If performance matters to you, you have to test.
Here is a small excerpt from my book High-Performance SAS Coding
http://www.amazon.com/High-Performance-SAS-Coding-Christian-Graffeuille/dp/1514362317
fyi and to use as your testing framework.
Data Set Buffer Case Studies
Each case is different and ideally tests should be conducted in order to decide upon the values of BUFSIZE and BUFNO as well as to decide when to use direct I/O. These decisions will depend on the hardware, the data, how the data is used, and what other jobs are running. [..]
The first row uses the default SAS settings. The best times are underlined¸ and the grey zones are where enabling SGIO slowed down the data transfer compared to just using the default.

May the "search for better performance" bug be with you forever! 