At the moment, the cache coordination feature is only available in Oracle TopLink. There has been some planning discussion related to how this kind of feature would make it into the JPA Reference Implementation, but I cannot tell you anything firm at the moment.
Depending on your definition, "100% safe and predictable behavior" is acheivable with just the locking strategies detailed above. A properly defined locking scheme ensures you never write incorrect data to the database and that instead, the system notifies you when you try to do that and gives you the opportunity to refresh data that is no longer up to date. The cache coordination feature does not necessarily guarantee you will never run into locks, it just reduces the frequency of them.
[Message sent by forum member 'tware' (tware)]
http://forums.java.net/jive/thread.jspa?messageID=236786