Text analysis doesn't seem to learn from previousy flagged spam messages

Issue Summary

Maybe I'm doing something foolish but I don't think so. Let's say I'm receiving numerous spam messages with the following text: "michael kors". I have several choices:

• I can choose the text analysis to report those spams. Mollom servers should then consider the pattern as spam after some time
• I can add the string "michael kors" to my own blacklist
• I can do both, obviously

However it seems that none of this work ok and I'm receiving more and more spams with the same strings and I'm flagging them endlessly. I'm reluctant to switch to the CAPTCHA because it has to come with a proper caching strategy (IIRC cache will be invalidated when a CAPTCHA is displayed)

Is this a known issue or only me?

[EDIT] I realize I didn't gave much info here. For the text analysis I choose to Automatically discard the post and I'm doing the following with the blacklist:

CONTEXT: all fields
MATCHES: contains