Taking on a new dba role for an Oracle 11gR2 2 node RAC, I've been told that about 9 times a year, in schedule with other software releases, we intentionally shutdown and restart the database. It causes all sorts of havoc - we have unbalanced nodes and jobs that ran fine before end up on a different node running poorly, re-caching of everything, etc. The database struggles for a few days as we put out new fires.
When I inquire as to why, it 'helped in the past' but nobody knows exactly why. Is there a valid reason? Is there a known memory leak somewhere in the Oracle RAC configuration that makes this needed? (That would be shocking.) A similar question for MS SQL Server indicates 'no', but Oracle is generally quite a bit different.