We describe an optimised consequence-based procedure for classification of ontologies expressed in a polynomial fragment ELHR+ of the OWL 2 EL profile. A distinguishing property of our procedure is that it can take advantage of multiple processors/cores, which increasingly prevail in computer systems. Our solution is based on a variant of the ‘given clause’ saturation algorithm for first-order theorem proving, where we assign derived axioms to ‘contexts’ within which they can be used and which can be processed independently.We describe an implementation of our procedure within the Java-based reasoner ELK. Our implementation is light-weight in the sense that an overhead of managing concurrent computations is minimal. This is achieved by employing lock-free data structures and operations such as ‘compare-and-swap’. We report on preliminary experimental results demonstrating a substantial speedup of ontology classification on multi-core systems. In particular, one of the largest and widely-used medical ontologies SNOMED CT can be classified in as little as 5 seconds.