Examples from the first 100 people in my database:
http://www.imdb.com/media/nm0946148/ 'Weird Al' Yankovic ... thumbnailed main image, larger images in list
http://www.imdb.com/name/nm0001173/ Aaron Eckhart ... thumbnailed main image, larger images in list
http://www.imdb.com/name/nm1594381/ Aaron Lazar ... thumbnailed main image, larger images in list
http://www.imdb.com/name/nm0479903/ Aaron Michael Lacey ... thumbnailed main image, larger images in list
http://www.imdb.com/name/nm0005455/ Aaron Spelling ... thumbnailed main image, larger images in list
Count of "good" images links ... 12
Total images ... 17
That pretty much matches up with the database overall...about 20% of the people have images, and about 30% of those don't follow the methodology you outlined.
I don't think this, nor the frequent page format changes are coincidental.
Here's an example of the 2 identical images with the secondary image being the 'big one'....and there's quite a few of these too...
http://www.imdb.com/name/nm0409691/The only way I can see IMDB large images being used is to have the program look at the full-image size and do a random download from the list if it's a thumb, and even then you'd need a 'reload / re-randomize' function when the image doesn't work out.
Meanwhile, 12% doesn't work for me. It's a start, but nothing more. From my testing, I can get that up to about 50% via internet downloads. This doesn't resolve which "john smith" I'm loading into my database, so screen shots are the most reliable way. With copy / paste from screen shot to people image the process could be reasonably effecient.
Mostly I collect older B-movies so the screen shot process is much more important to me than people who collect the current releases. I have no doubt some people will have 50% of their cast filled with images via IMDB...thumbs or not.