[Mimedefang] Subject block - good success - Anyone else trying it? [REDUX]

[si at yacc.co.uk]

Way back in July 2005 there was some discussion re- wisdom of
maintaining a database of spammy subject lines, and blocking on the
basis of this.

Consensus seems to have been that, in the majority of cases, Bayes was a
better bet.

Recently (over the last year or so) we've been seeing an increase in
message size to well beyond 1MB, which for us meant that mail didn't get
scanned by SA, and hence went unnoticed by Bayes. We did the obvious
things, including increase size of message passed to SA, etc.

We've been experimenting with a 'spammy subject database' too.
Essentially, if we have a 'big' message and subject is in the database,
we pass to SA, otherwise we bypass SA. 

There are a few downsides to this approach (slightly lopsided Bayes
learning for example), but seems to be a nice compromise so far.
Question here is do we spend the time and effort to produce something
production quality or not?

Just wondered if anyone else is doing anything like this, and if so,
what kind of results you're seeing?

I'd also love to hear from anyone rejecting outright on basis of subject
alone (as proposed in the original 2005 post). 

Thanks for your time

Mup.