SpamAssassin Rulesets - add in rules by Jennifer Wheeler.  Props to Adam Lopresto and my SARE buddies. 

backhair.cf  (1.6 : 2004-04-05) --[words obfuscated by nonsense html tags]  Y<oivugriub>ou cou<iuqgheriugv9h>ld rea<y>lly u<owiuer88>se this.

chickenpox.cf (1.18 : 2004-04-05) --[words obfuscated by non word characters] Th1s |s a v3ry h4ndy se7 t0 c@tch th!s 50rt 0f (rap.

mangled.cf  (1.60 : 2004-06-04) --[Set of obfu/gappy words. exclusions allow for higher scores to bump up sleeper spam.]

weeds.cf  (1.1 : 2004-04-05) --[looks for Alpha decimal/hex encoding, lower and uppercase. Use this or Weeds 2, not both]

weeds_2.cf  (1.1 : 2004-04-05) --[same as the above Weeds set, but more limiting. this set is "safer", though I find the above set safe for my tastes. Use this or Weeds, not both]

The only set I would be caution you with is mangled.  If you're conservative, pull out the rules for smaller words (4 letter words or less) because these have more of a tendency to fp. As with any rule you didn't write, score each rule based on your setup.  Set them low, see how they hit, and adjust accordingly.  You could also go through each rule and decide which words you choose to target, and pull the others.  Another option is to drop scores on smaller words.  I expect these to hit a bit in ham, but the idea is, there shouldn't generally be enough other junk in the ham to make the entire message FP...  just juggle like you do everything else until it works for you.  The reason I made this set is for very generic well written spam that might need a bit of a nudge to eek it over the top.