This could be possible with a script, IF the results of the Text Recognition process are reliable AND the text always appears at the same spot (more or less) on the page. The implementation is a bit tricky, though. You would need to examine the quads of each word on the page and compare it to the area where the page numbers are supposed to be. If a match if found you add this page to an array that holds the original page number and the new page number. If no match is found then you need to decide what to do with those pages.
Then you can either move the pages around using a sorting algorithm to the new order, or (my preferred approach) extract them as individual pages and then re-combine in the correct order.
As I said, this is not a simple task. If you're interested in hiring someone to develop it for you feel free to contact me privately at try6767 at gmail.com.
Thanks for your input. Interesting, the numbers are all in the same place.
I wonder if it's possible to automate bookmarks for each page based on these numbers and then sort numerically by bookmark... end result would be the same.
It's probably not worth anything to me monetarily, as it is I can perform a search for the document by number and find what I need. I'm trying to stay away from manually sorting, just trying to clean up the stack as it were...