Altogether, the dumps are 650GB uncompressed, 40GB compressed.
I think a reasonable next step for this project would be to write filter
scripts that take a compressed dump, reduce the article count in some way,
and then recompress it, possibly in a different format. For instance, we
could have a "most popular 4GB" of the English Wikipedia, based on page
view statistics, recompressed as an SQLite database.