In gmail I guess that once a few peeps click "Report Spam" on a mail that passed filters, similar mails are also flagged as spam in other accounts. I've always found gmail extremely effective at this, and I practically never flag mail as spam myself.
Sure, there are automated spam filters to configure, but overall wouldn't going alone make things much less efficient on that topic?
EDIT: Greylisting would probably cut down on this further drastically, but I'm just too lazy to implement because no one complained yet about too much spam. :)
Anyway, just did a quick search through mail.log and in the past 24 hours, I got 4017 SPAMS that were rejected at SMTP time by Postfix , about 40 SPAMs got through those, but caught by SpamAssasin . 0 spam delivered to inbox. 0 false positives. 126 legitimate mails.
I'd looked into greylisting some years ago, but never implemented because I didn't like the idea of delivery delays.
So I'm in complete agreement with you regarding how effective SMTP level checks + rbl are, but I disagree regarding statistical filtering -- Bayes was key for me to getting 100%. I had put off adding this layer for years, because I didn't fully understand or trust it, but now I'm a believer. Also, once trained, it just works. I rarely look at my spam folder. Retraining is easy, I just drag a rare false positive/negative to HamCaught/SpamMissed folders then nightly training script auto learns.
 I was an email admin at a local gov agency, then tried to get into email hosting biz, so it wasn't just personal obsession.
 Using things like standard helo/client/sender/recipient restrictions, with very few custom rules, and rbl check using spamhaus.
 SA + razor/pyzor/bayes, again with very few custom rules.