Hi Richard
This is just my wild guess, but my guess is that you have to look at
the larger picture. As Susan said, "many grains of sand" ... assume
every grain of sand is a signal that is sent by your website. CSS
image replacement is just a tiny bit more than hiding content on your
page (except that the place where the hidden content should be is
filled with an image or similar). Hidden content through CSS is mostly
easy to recognize algorithmically. That's a grain of sand. If you're
replacing headers, that's probably another grain of sand (aka
"signal"). If your javascript does strange redirects, that's some
more, if your pages uses 10 lines of alt-text for a 1x1 pixel image
that's probably some more.

If you have enough grains of sand, if the Googlebot brings his beach-
towel when visiting your site, then chances are you've either gone too
far or things are being misinterpreted.

My guess is that there is a threshold of "sand" that brings a site
into a queue for a manual review. This is where intent and replacing
content with the same content in an image comes along. If things look
good, no problem, your personal threshold might get adjusted or things
might get updated in the algorithm, etc. This kind of review is
probably rare and I bet the queue is pretty long.

However, there is likely also a threshold where the Googlebot brings a
bulldozer instead of a beach-towel. That much sand just can't be an
accident. With that many signals being sent, it can be assumed that
the site is in fact doing something wrong on purpose (and I bet most
of the time they're right). These sites are likely to be penalized
automatically.

It's a bit like my email spam-filter works: it assigns a score based
on many, many rules. Below a certain number is ok, within a range
slightly above it's placed in a check-these queue, with a score way
above that it's automatically discarded.

I'll assume that the automatic penalties are pretty much ok, if you
have that many spam-signals in your site, then you're either doing a
whole lot of things wrong on accident or you're doing it on purpose.
The middle range is a bit harder: are they penalized with a pseudo-
penalty or are they really manually queued?

Also, which kind of items cause how many grains of sand? Hiding text
with CSS? Content duplicated from other sites? Sneaky javascript?
Stuffed headers? Link-exchanges? Links to strange sites? Interesting
things!!

John