Personally, I'm not a big fan of genotipic costs - that is, costs associated with DNA operators - or with costs that are associated with meta aspects of the DNA such as it's length or the number of bps executed, etc. Such costs provide selective pressure, but towards minimalism I.e. shorter genomes or the execution of the miunimal number of base pairs or minimal use of specific base pairs.
I much prefer morphological and environmental costs - that is, costs that are associated with what the bot actually is or does, not on the code it uses to do it. This subject touches on the separation between genotype and pheneotype, something that is half pregnant in DB. In biology, having more DNA is basically free. Even 'executing' more DNA is basically free given the natrual world's inherent parallelism. What matters is what the end result is. Larger bodies/beaks/penises has a real cost in the real world. What specific DNA was executed to create those larger brains/trunks/fins can bascially be ignored as far as costs go. Inheritability and evolvability though, well, those do influence DNA structure and exhert selection pressures, but that is different from what IMHO should be purely morphological costs...