This is an interesting idea.
How do you deal with any PDF or image where the text is not OCRed? Perhaps I need to try this approach and setup a smart rule that OCRs newly added image or PDF so that at least the text content of the item will be available for search.