The configuration I've pasted below worked for me, though it requires the page numbers to be in exactly the same place on each document.  A more robust solution for getting the page numbers would be using a large zone + regular expressions (to allow for the page or numbers being shifted a little).

One subtle thing about Last Page Identification is that when you have it configured like this, the First Page Identification steps are skipped until the current, identified document is complete (by satisfying the Last Page Identification conditions).  Also, if a page limit is set, that will be enforced before Last Page Identification is run and can also "complete" a document.

Also, the tokens from other processing steps (e.g., Page Processing) are not available in Last Page Identification--I don't recall the reason for this, but it appears to be the case.  However, if you've performed full page OCR on the document in a previous step (e.g., Pre-Classification Processing or Page Processing) and configure your Zone OCR to "Use existing text", it will use the OCR results and not re-OCR that part of the document (for a minor speed boost).