I like the idea of having the fumble rate depend on skills, and I see it as a fine, so-far unexplored area to investigate with the objective of finding new house rules. Ultimately, these house rules can be used to show the "danger level" of certain parts of a setting.
My thoughts on this area are that the relation between fumble rate and skill proficiency should be stronger and more varied than it is now. The above suggestions only discuss the lowering of fumble rates, but if the current rates are average rates for everyone in the setting, such approaches will unbalance the system. What I think would be useful instead is to assign a minimum (lower than current value), a maximum fumble rate (higher than current value), and an algorithm for decreasing the fumble rate for each attack table.
Assigning the limit values should be reasonably easy, whereas the algorithm is a bit more difficult. It could well be argued that for unlearned skills - in addition to the skill penalty suffered - the fumble rate must be quite high. I think it would be reasonable to merely double the maximum fumble rate value.
Regarding the algorithm, I think that the decrease in fumble rate should be diminishing, requiring more and more ranks for a decrease. Something like requiring the same amount of ranks again for the rate to decrease (1, 2, 4, 8, ...). I'm sure somebody can come up with the right mathematical formula for calculating the decrease in the fumble rate...
For weapons, a Dagger (normal FR: 01) could have a range like 1 to 4 (denoted [01, 04]). This would mean that a person with 0 ranks would have an FR of 08, with 1 rank the FR would be 04, with 2 ranks it would be 03, with 3 or 4 ranks it would be 02, and with 5 ranks or more it would be 01. For a Broadsword (normally FR: 03) the range could be [02, 07], which would give the currently "normal" FR at 9 ranks.
For spells, fumble rates should only be applicable to spell with attack tables; the casting of the spell should be handled differently. To the attack spells, this approach offers the opportunity of graduating the difficulty for different kinds of spells, for example by letting Arcane spells have a wider span of fumble rates. This also has the potential of making the casting of spells much more dangerous for an untrained caster.
In fact, lots of opportunities for adjusting fumble rates could be introduced, for example a spell focus that decreases the FR by some number or magical herbs that do the same. And similarly for weapons, a sword with a better balance or a lighter mace could decrease the FR, while more complex weapons could decrease it.
In any case, relating fumble rates to number of skill ranks could also be an incitement to develop skills that gain only a small bonus increase per rank. And I think it expresses the difference in proficiency between characters of lower levels compared to those of higher levels.
A final thought is about category ranks. While I think that it would make sense to add these to the number of ranks in a certain weapon skill, the same can not be done for spells, as their categories cannot be developed. Therefore, I think they should be left out of the equation so as not to get two different algorithms. Another option is to rework the rank bonus system of spells (from "special" to "standard" progression), but that's an entirely different matter!
//K