All clustering subsystems (and some non-clustering subsystems) use a Schema enumeration, and register a subsystem parser for each schema version. Since only one subsystem parser will ever be used, and only at startup, these should be created via a factory so that the parser can be garbage collected after initial startup.
For those subsystems that use PersistentResourceXMLDescription, only the parsers for legacy version should only be created on demand, since the PersistentResourceXMLDescription instance for the current version was already created for use by the subsystem writer.
Similarly, parsers for deployment descriptor schema versions should also be created on demand.