State of the art reasoners for expressive description logics, such as those that underpin the OWL ontology language, are typically based on highly optimized implementations of (hyper)tableau algorithms. Despite numerous optimizations, certain ontologies encountered in practice still pose significant chalo such reasoners, mainly because of the size of the model abstractions that they construct. To address this problem, we propose a new blocking technique that tries to identify and halt redundant construction at a much earlier stage than standard blocking techniques. An evaluation of a prototypical implementation in the HermiT reasoner shows that our technique can dramatically reduce the constructed model abstractions and reduce reasoning time.