I stand by my argument.
Mutations to junk DNA can still be deleterious as occasionally a mutation will enable some new hunk of DNA via the insertion of a start command or the deletion of a stop command. The longer the genome, the higher the probability of this happening. One is then left with the question that when it does happen, is the result benificial or deleterious? Most changes of any sort, including the sudden expression of new logic, are deleterious - stores to sysvars get overridden by the new code and so on. Additionally, point mutations are more common than indel events. Thus, noncoding junk DNA within a coding sequence (such as boolean operators within a start-stop block) represent increased surface area for mutations that can turn non-coding sequences into coding sequences. Thus, in my opinion, having a longer genome - even if all that additional length is noncoding - all else being equal increases the chances of a deleterious mutation and therefor will be slightly selected against.
I actually see this in my sims quite often, but (and this is key) only when there are no viruses. Over time, the DNA becomes more compact for the same function. Non-coding interons such as boolean operators outside conditional blocks get selected against. But as I mention above, if there are any viruses in the sim, my argument does not hold or rather, it becomes moot. The selection pressure favoring virus propagation utterly and completely swamps any downward selection pressure on bot genome length. I suspect that most people who witness increasing DNA lengths in their evo sims are seeign the effects of viruses. I am actually to the point of wondering whether given the current mutation logic and default settings, anything but the most modest increases in DNA length are even possible in reasonable time frames without virus involvement.
By way of example, below is an evolved zerobot from my long running sim now at 21.4M cycles and 1250 hours run time. The organism in question is of generation 2681 and is the product of 514 mutations from the starting zerobot organism which consisted of 30 0's. After I posted my sim at 1000 hours, I made some slight changes to settings and veggies such that they died out faster than the mean virus infection rate, which effectively removed them as viable virus propagation hosts. Over time, the virus that had evolved in the sim died out. Since then, there has been downward selection pressure on genome length even though I have no DNA costs. The average genome length in the sim is now 57 base pairs (the length of the guy below) down from around 130 when I posted the 1000 hour sim. Note in the DNA below that start is now the first base pair. Remember, the first ancestor was a pure zerobot, no start command. This guy's more recent ancestors had junk ahead of that first start but there has been selection pressure to remove all non-coding crap prior to the start and elsewhere. The DNA is actually starting to look more and more like hand coded DNA with regard to it's structure. Isn't that interesting?
start
%=
=
pyth & dec
1 .setaim store
*.dx ~ -8 >
rnd ++ 1 1 xor
*566 -800 dist -25 !%=
start
not
1 1 dup 38 56 pow << store
and
4 ^ ^ or
21 dist mult store
store
pyth 12 *.robage * store
*.aimright *.shootval ~ store
917 xor
store
div
With respect to the question on biological seleection pressures on genome length, I'm no biologist, but there must be some morphological carrying cost associated with longer DNA. DNA lengths are not infinite. For one thing, longer DNA means you need more necleotides to divide, more raw material. I can imagine it might impact rates of transcription, impactign cell function effeciency and perhaps have a large effect in species where reproduction rates are critical such as bacteria or viruses. So, its just personal opinion, but I woudl expect that all else being equal (which it never is) we should see genome lengths in nature reflect a similar balace of selection pressures.