You're right, the insertion cost equation: a + bx does NOT
make much sense. Gonnet, Cohen, and Benner (1992. Science
256:1443-1445) in their "Exhaustive Matching of the Entire Protein
Sequence Database" have found that the probability of a gap occurring
in an alignment of two sequences increases linearly with the PAM
distance separating them and that the following equation, where P is
the probability fo a gap of length k, fits the data accurately.
10log(P) = -36.31 + 7.44log(PAM distance) -14.93log(k).
It's an interesting (and exciting!) paper.
Good luck. Edwin Rock