Extensibility of a data language

It seems that most successful examples of extension exist at the level
of protocol, display languages, and so forth, whereas I can't think of
any particularly successful data language where the language had the
possibility of extension of the language by users built in.

It often seems that Data heads feel that the kind of loose
extensibility that namespaces can be set to allow will create anarchy.
Any arguments against?

Cheers,
Bryan Rasmussen