I think caching is something that Amazon (and probably others) rules out explicitly. The point of that service is to gain hits, sales and user info, not providing a nice source to harvest for your own server.
It's not about caching thumbnails themselves, but URLs to them. There is no interest in caching images when Amazon or Google have CDN designed to deliver their content in a breeze. So the number of hits to the images stay the same for the providers, originating from browsers, whereas the number of requests to AWS or Google Book services decrease, since their result is cached. And this cache may be cleared periodically, or checked since providers can decide to move their thumbnails URLs. I described a solution for bug 7187: http://bugs.koha-community.org/bugzilla3/show_bug.cgi?id=7187 And indeed, I just reinvented the wheel: http://wiki.koha-community.org/wiki/Covers_from_multiple_sources_RFC