While profiling a heavy-load scenario using a database, we've discovered a piece of code that for new nodes (i.e. nodes added via parent.addNode) always hits the Infinispan cache store, in an attempt to search for the node.

This has a *huge* performance penalty in a case when we know up-front that the node will *never be* in the cache.

After fixing this, profiling the code shows *significant* performance improvements.