An object database management system (OODBMS) has been often criticized for its alleged insufficient scalability for a large-scale production system. We investigated the scalability issue on a commercial OODBMS with a focus on the scalability with respect to the number of objects. Our approach was a benchmark experiment using the loading and indexing of SGML text documents as an application. The application was characterized by its small granularity of objects, which resulted in a huge number of objects in order to make a large database volume. The OODBMS we used was built in a so-called "object descriptor architecture (ODA)" as opposed to a "virtual memory mapping architecture (VMMA)". The results showed that the OODBMS scaled better than we had anticipated. It required, however, algorithmic resolutions to overcome the shortage of object cache space. Three key resolutions were made. First, we created indexes in fragments by committing a loading transaction before the object cache space became full, and subsequently merged the fragments into one master index. Secondly, we had the application release cached object descriptors (CODs) as soon as they became unnecessary. Thirdly, we utilized a query cursor mechanism to fetch the objects returned from a query piece by piece without overflowing the object cache space. Currently we are attempting to push the scalability up to filling up the maximum available hard disk space.