Non-standard database systems become available now, even as commercial products. They overcome a lot of deficiencies of relational systems w.r.t. their use in engineering applications like computer-aided design or software engineering. Their rather sophisticated functionality especially concerning the manipulation of complex objects makes them highly attractive for engineering applications. If being used as the central database of a rather complex application they could however still become a bottleneck w.r.t. performance. This paper presents a new way how to define a special purpose benchmark which enables to select the fastest database system for a particular software engineering application. It is argued that existing benchmarks are not appropriate to support such a selection, because they neglect important application specific characteristics which significantly influence the database performance.