Is still denormalisation but one big advantage of not just augmenting the database meta data is that you can describe types without needing actual database tables. It is still re-inventing the wheel but it is hard to sell having lots of tables in the database even though the database is very able to cope. Tooling may struggle opening 1000s of tables in the UI.
e.g. You want to describe a model and user interface but never want to collect the data in a table. So you want the model definition but no need for the actual table (if not using the meta data).
I think I'd still vote for using real tables but would need to test that all parts of the database and tooling can cope.
