They should just put the stuff they extract from the PDF into some queryable form. Then, when there is a hit on a query, direct to the PDF. I doubt that the query results will be significantly worse, probably even better.