Recently had a similar challenge. You might try checking out IBM's site by searching for xml parsing techniques. My files were huge and the solution I worked out performs extremely well and quickly.

Timoteui