In my experience (being admin of the V3C board), the 'silly question' method is integral part of the 'bulletproof spam wall'.
Another part is the software we use, buying an owned licence for vBulletin cost us (V3C) around 150 pounds, but even while I haven't done anything substantial to it in over 3 years, we still have a near-spamfree board.
Only action needed was to (slightly) alter the silly questions as soon as the first spam registrations started to show up.
In the old days (running a CGI/perl board software called Discus, which was quite Discus(ting) indeed!) new registrations were filtered through mod approval, which works well as long as you don't think too much about the mod workload.
After the initial transfer to vBulletin, I have experimented with combinations of the existing mod approval with out-of-the-box "protection mechanisms" like CAPTCHA, limited posting rights for newbies, etc, but only after I tried the silly questions all spam registrations stopped and mod approval became superfluous. With CAPTCHA (any implementation) quite a lot of spammers still got through. You know how that looks on the forums pages...
Due to developments in both image analysis software and mass-production sites where the codes are typed over by low-paid humans, CAPTCHA has become even more useless over time, and today I wouldn't advise anyone to use it for any purpose, unless you modify the parameters like the TTF fonts and result image size, character set of the guessword, etc to a quite heavy extent.
All of which makes human registration harder, too... Catch 22 is the same as silly questions, but more negative.
The main problem with the silly question idea is that it is language specific. On V3C both the questions and the required answers are in dutch. I have no doubt this is one of the prime reasons we have been spam-dry for so long with so little mod intervention.
It seems the challenge for Aymat is to find the right questions/answers that are specific enough to deter spammers with some English language skills, while still easy enough for interested human beings to answer correctly.
you could start a competition to think of these questions?
