Bimodal Distribution of Mismatch Frequencies in Illumina

[mdjones66]

I created a pileup file using Maq. If I look at the log transformed distributions of frequencies of mismatch for each nucleotide across all positions,

(forward mismatch /forward depth + reverse mismatch/reverse depth)/2

, I end up getting a bimodal distribution for each of the nucleotides accept one. I think 'A' looks more normally distributed but I can't remember for sure. Has anyone else seen this and do you have any idea what could be causing this?