And of course it would be trivial to serve the Googlebot an innocuous
CSS but the same could be said for serving the bot anything innocuous
while serving everyone else something different.  CSS, HTML,
Javascript, wouldn't really matter.

I could think of a trivial brute-force method for determining to a
certain extent what CSS does, render the page and then run an OCR over
it.

I'm not saying how likely it is that Google does that but where there
is a trivial brute-force method thought up by a Neanderthal, like me,
there is usually a more elegant and efficient method possible to
someone who knows what they are doing, e.g. Google-heads.

Craig