Here's my 2 Eurocents.
1. Doubles method. Our Cthulhu GM uses this at the moment. Success levels are compared via "blackjack", ie. higher is better, as long as it's a success.
I can see the benefit of this method, but I still don't like it too much. Special results happen too rarely, and the chance of a fumble is generally too high.
2. One-tenth method (as per MRQ). I like the simplicity, but special results still happen too rarely, as above.
3. RAW, with different special and critical results. Used this for ages in RQ3 and got used to it, bt I think the calculations are a pain in the butt.
4. My houserule: I only use special successes (one-fifth skill), and no criticals (for the sake of simplicity). For weapon skills - which are rolled most often - there's a "special" space on the character sheet, so there's no math. Maths. Whatever.
There is no comparison of success margins except for "special beats normal success", because I don't need it.
Fumbles only happen on a roll of 99-00, until your skill reaches 98+, at which point you only fail on a 99 and fumble on a 00. (I think I stole that one from Elric/SB.)
Granularity may be nice, but I like simplicity better. YMMV.