[Spambayes] many problems/question on spambayes

[es at javanet.info]

Hello all,
I receive a lot of spam (1 each 2-3 min that pass SpamAssasin checks) 
and I've trained SpamBayes with 6253 spam and 1764 ham and it correctly 
identify (only) around 80% of spam. However, it act in a weird way, 
because looks like it completely skip to check some (5%) mails (no 
header is added at all or I even saw one with just the 
X-Spambayes-MailId added).
Another problem is that I want the ones with probability > 0.7 to be 
classified as spam as they are all spam and I didn't saw any option to 
fine tune this. Also the scoring/evidence is low for common words like 
'buying': 0.35; and would be nice to let the user act in someway to 
modify the scoring for some words.
Another mystery is why thunderbird filter on X-Spambayes-Classification 
"contains" "spam" miss some emails that has been correctly tagged as 
such and unfortunately I can't set the filter to X-Spambayes-Probability 
 > 0.7 because with custom header filters there is no option to check 
for numbers like is possible for the dimension...
I can try with sb_filter/procmail but I haven't tried yet because of my 
fairly complicated qmail+vpopmail setup and first SB should give 
reliable result already with the imap filter and I was wondering why 
there is no option with sb_imapfilter to actually take an action after 
the email has been classified (like delete/move etc). This looks very 
logic to me to be the job of a filter... why this should happens down 
the line relying to the (poor) email client features? Very few webmail, 
for example, has options to classify on custom headers. I would like 
your opinion about this.
My test has been made with spambayes-1.1_alpha3 available trough 
gentoo-portage don't know if the alpha4 has some important bug fix for 
my problems.
thanks for any advice...
Eli