I understand the problem and will consider ways of providing storage-optimised data models in future releases. The additional title and namespace information is relevant for implementing filtering operations more effiently (i.e. without an additional join with the page table) and hence was denormalised on purpose.

For relations and objects, ids are not always available and thus cannot replace the title strings in general. We could consider having an own indexing scheme for them or use ids at least for attributes (which need to have articles before being stored). However, ids in MediaWiki are not as persistent as the name: when you move a page, the id of the page of the original name changes. Hence all uses of ids require global updating operations whenever pages are moved and we also tried to limit this.

What you can do to save space (at the expense of performance) is to drop indexes which are currently built for the tables. Especially the ones over object and subject titles/namespaces are not required for all operations and deleting them might be feasible.

The new storage engine of SMW uses internal numerical ids for all pages, so that title strings vanish from relation tables and subject positions of other tables completely. In compensation, there is a new SMW-specific id management (we cannot use MediaWiki ids since they do not exist for all objects occuring in SMW tables). Yet, for relation-heavy wikis, this should be a great reduction of storage space. The new storage implementation is in SVN and will also soon be released with proper update instructions. For testing it now, see the instructions in Bug 13960 (switching back to the old implementation is always possible, but you should run on current stable release SMW1.1.1 before trying out SVN.