In the specification for OWL 2, the section on key axioms states:
The semantics of key axioms is specific in that these axioms apply only to individuals explicitly introduced in the ontology by name, and not to unnamed individuals (i.e., the individuals whose existence is implied by existential quantification). This makes key axioms equivalent to a variant of DL-safe rules. Thus, key axioms will typically not affect class-based inferences such as the computation of the subsumption hierarchy, but they will play a role in answering queries about individuals.
It then proceeds to show an example of a set of axioms that would obviously entail a conclusion but that the spec does not recognize, because of that rule.
What are the reasons behind this decision?