In mammals CpG dinucleotides (the targets of promoter methylation) are hypermutable (they tend to mutate to TpG). This seems an odd feature for a DNA holoenzyme to pick up; my suspicion is this is a slow mechanism for making epigenetic modifications permanent, allowing faster, more flexible evolution of gene expression. A promoter that tends to be frequently methylated is more likely to mutate and become permanently less active.
