Ian Bicking writes in a comment to an earlier post about "issues" with certain anachronistic databases (e.g. ZODB and Berkeley DB)...
RDBMSs and flat files are the two true ways, because they are somewhat less broken than other things!A Smalltalk image works well as a simple object database because it is fairly simple: in-memory, simple process model (e.g. no complicated conflict resolution). Prevayler could become a simple extension of this, and the capability/complexity trade-off grows from there.
I agree with Ian's sentiment. The relational model is better than credit is given from the object-oriented programming community. The main problem is not the relational model per se, but the language implementation (i.e. SQL) is not so compatible with the more modern languages we use (even Smalltalk and Lisp, which pre-date SQL by a fair number of years. I agree with other voices that the Object-Relational Mapping problem is not worth spending too much time worrying about.
Even with SQL's shortcomings, the range of reliable implementations is nothing to shake a stick at. And flat files in basic file systems are tried and true, not a lot of moving parts to worry about for a good many scenarios.
Newer implementations of relational databases do address some of the language issues. (e.g. Derby nee Cloudscape and the even simpler Hypersonic SQL on the language issue, Postgres on the some of the O-R mapping issues). But some down to the roots rethinking of the implementation of relational databases could fuel them up for the years ahead.