Brian,
I've never experienced anything like this myself, but you mention that the missing areas have three-fold coverage whereas most of the remainder has perhaps only two-fold coverage. I wonder if issues with the original images mean that, where there are three sources, the results are so divergent that PhotoScan is rejecting the tie points. In absence of any better suggestions, I wonder if you re-processed and eliminated one of the images there, what the result might be?
Dave