Structuring rasters?

[Jeff Dege]

We have a customer who wants to integrate aerial photos into their maps.

The extent of the map is that of a middle-sized state.  200-400
kilometers.

The tightest zoom is a couple of blocks.  100-200 meters.

(Map size scales with the browser - figure 600-800 pixels.)

I've done a bit of playing around with the GDAL tools with a
county-level dataset.  From this I learned that:

1.  You can futz around with this stuff forever, never getting it quite
right.
2.  You're chunking enough data that every failed attempt takes forever.

The question is, given a few tens of thousands of hi-res jpegs, what is
the best way to restructure them so as to give acceptable performance
through Mapserver?