The process of producing the maps and delivering them to Paizo usually involves flattening the text layers into the image, so the numbers and labels are part of the image itself. This makes the files a lot more manageable in terms of size: they transfer to Paizo more smoothly.
That said, I've started providing them with layered versions where the text can be edited at Paizo if they find a typo, so that may open up the possibility for something like the PDF image capture you're talking about. The text is still in Photoshop when I send it to them, though, with special layer adjustments to create outlines, shadowing, etc--so I don't know how hard it would be for them to remake it in their design & layout software.
