"Doctor, it hurts when I raise my arm like this." "Then don't raise your arm like that."

You have GB's of data ... and you want it in a single XML file? That's gonna suck no matter what you do with it. Perhaps you should use some sort of database, whether it's GDBM or some RDBMS (DB2, Oracle, MSSQL, MySQL, PostgreSQL, etc.). It'll be easier to build and easier to access...

(If it's hierarchical data rather than relational, I think DB2 v9 has the ability to store the XML data as hierarchical data - "PureXML"(TM) - making it simpler to use without changing your algorithms very much. I imagine that most other DBMS's either have something similar, or will have something similar Real Soon Now(TM).)