Tamar Weinberg at Search Engine Roundtable reports  that in a Google Groups forum, a Webmaster Central team member stated that you could use something like the z-index attribute in DHTML styles to hide text or links behind an image, so long as the text/link being hidden is what’s represented in the image.
I think it’s a good thing that they do allow this sort of use, because it appears to me that they’re doing this very thing on their own search results pages! If you refresh a search page, you can see what they’re hiding under their own logo:
…a text link pointing to their homepage.
Now, the interesting question I’d have for the Google team about this would be: this is straightforward if the image itself contains text, but what would be allowable if the image doesn’t contain text, but say, an image of a lion? There’s many different ways to express what that lion is from “lion” to “tawny, golden-furred lion king”.
Or, should we be assuming that images that are written over text and links are only allowable when the image contains text?
The Google Webmaster Tools contributor states that you could be using image’s ALT and TITLE attributes to essentially do the same thing. This is sorta funny, because one could say the same thing of Google’s use of this on their own page — why are they doing it?
One immediately wonders how Google polices this, since they’re apparently not frowning upon pages drawing images over text/links in all cases. They can detect text written over images, but would they have every instance checked by a human? Or, are they using optical character recognition algos to automatically check the text within images against the text being hidden?
In any case, the fact that Google is doing this on their own site could be taken as more confirmation that they don’t consider the technique to be bad in of itself — as long as the practice is conservative and the text/link just describes the text content within the image.