This is a good question. Yes we have adopted a similar idea. The forward chaining based inference engine in Oracle Database will only bind existing known RDF resources to variables in rules. We will not generate a new RDF resource that does not previously exist in the ontologies. This is very important because it ensures the inference process will finish in a finite amount of time.
The above does not apply to user defined inferencing (a 12.1 new feature) though.
Thanks a lot for the answer.