Google could technically do a similarity measure between images coming from similar links (or similarly worded emails) and then provide the cached version. But

Google could technically do a similarity measure between images coming from similar links (or similarly worded emails) and then provide the cached version. But then advertisers could hide random information in the image (steganography), making two visually similar looking images but with dissimilar enough content, that Google will be have to up the ante. And so on and so forth. But you are spot on them being "willfully dishonest". The way I look at it, they are at this point trying to push the barrier and see where users will protest enough that they need to roll-back.