Originally Posted by Varyon
It gets a bit less messy if you're willing to (le gasp!) use nonstandard dice. Rolling 3dF (dF is a d6 that has 2 +'s, 2 's, and 2 blank faces rather than numbers) would give results comparable to using the average (it has a max of +3 and minimum of 3, but going beyond that with averages likely calls for statistical outliers anyway), while avoiding the need for fractional skill levels. If there's an attack that is important to know when in the sequence it occurred (such as a Critical Failure that invalidated future attacks), you can get an approximation of when it happened with a 1d6 roll  a 1 means it happened on the first attack (tough luck), 2 means it happened 20% of the way in, 3 is 40%, and so forth, out to 6 meaning it happened on the last attack. You can get more finegrained with another 1d6 roll  2,1 would mean 20%, 2,2 would mean 24%, 2,3 would mean 28%, and so forth, out to 2,6 meaning 40%; then another 1d6 could break it down further. Of course, a better option would be to just roll d% (2 d10's, with one designated as the 1's place, one as the 10's place, and 00 treated as either 0% or 100%, depending on your inclination).
This does require a calculator at the table, of course.
