Roughly in order of planned implementation:

1. Data structures for "shadow trees," which would let you keep track of the order in which nodes appear in the document more efficiently.

2. More robust parsing.