Custom procedures are just Java methods annotated a specific way so the show up in cypher as callable procedures. These methods and all their supporting Java code is package in a jar for deployment. I am not sure what would constrain the number them. Anyway, I feel that if you built some custom rules engine in Java for use in Neo4j, you would not need 6000+ entry points to your engine. You would probably have a small set for configuring your rules engine and small number to interact with your rules engine.
What do you envision doing in each of these 6000+ procedures that they need to be callable from within cypher queries?