I decided to go for a database because the SGML file data structure is not the structure I want and also because I think a database would be faster for lookup (the next step in this script is converting the data to a hash of arrays so I have a hash lookup).

I'm not sure I understand your second point. Once I convert the SGML file into a database I don't use the SGML file any more.

Is there any reason to assume WinXP won't handle files of that size?