The actual details would be rather differen in PyPy, but we would like to have
the same optimization implemented.
+Or maybe not. We can also play around with the idea of using a single
+representation: as a byte string in utf-8. (This idea needs some extra logic
+for efficient indexing, like a cache.)
.. _`optimized unicode representation`: http://www.python.org/dev/peps/pep-0393/