Just because I can't run sa-learn on my 15,000+ messages spam folder. It'll crash due to some hardware problems.
I thought that splitting the mailbox into smaller files will allow me to feed it to sa-learn.
I'm not sure something similar doesn't exist but I wrote mine anyway ;-)
The only problem is it consumes a lot of CPU and RAM, it was killed/crashed multiple times but it worked and allowed me to feed my spam mailbox to spamassassin!
Here it is in case someone needs it: split_mailbox.py. Needs python 2.5