Good alternatives for massive deployment.

[Carlos Paz]

Hi!

I'm considering to deploy bogofilter for some 10K users on the server 
side, but I'm concerned about the disk requirements for all those 
training databases (some 10-15GB?).

I believe that most of the job could be done from a global wordlist, and 
users could personalize their filter training a smaller database, on the 
"errors" made by the bigger one (maybe they collect niggerian scams or 
want to block some particular messages not related to common spam).
Could this be done with bogofilter?
How could the two dbs be mixed statistically, achieving the desired 
behaviour?

Any ideas/advice are highly welcomed.