Obligate symbioses with nutrient-provisioning bacteria have originated often during animal evolution and have been key to the ecological diversification of many invertebrate groups. To date, genome sequences of insect nutritional symbionts have been restricted to a related cluster within Gammaproteobacteria and have revealed distinctive features, including extreme reduction, rapid evolution, and biased nucleotide composition. Using recently developed sequencing technologies, we show that Sulcia muelleri, a member of the Bacteroidetes, underwent similar genomic changes during coevolution with its sap-feeding insect host (sharpshooters) and the coresident symbiont Baumannia cicadellinicola (Gammaproteobacteria). At 245 kilobases, Sulcia’s genome is approximately one tenth of the smallest known Bacteroidetes genome and among the smallest for any cellular organism. Analysis of the coding capacities of Sulcia and Baumannia reveals striking complementarity in metabolic capabilities.

http://www.pnas.org/cgi/content/abstract/0708855104v1

I have no stake in this issue and (to my knowledge) no connection with the authors — it just caught my eye whilst I was browsing the arXiv’s quantitative biology RSS feed.

There’s at least one study that looked at the question at the level you suggest (including a competition experiment).

Genome Research 12: 408-413 (2002)

Domain-Level Differences in Microsatellite Distribution and Content Result from Different Relative Rates of Insertion and Deletion Mutations

David Metzgar, Li Liu, Christian Hansen, Kevin Dybvig, and Christopher Wills

Abstract
Microsatellites (short tandem polynucleotide repeats) are found throughout eukaryotic genomes at frequencies many orders of magnitude higher than the frequencies predicted to occur by chance. Most of these microsatellites appear to have evolved in a generally neutral manner. In contrast, microsatellites are generally absent from bacterial genomes except in locations where they provide adaptive functional variability, and these appear to have evolved under selection. We demonstrate a mutational bias towards deletion (repeat contraction) in a native chromosomal microsatellite of the bacterium Mycoplasma gallisepticum, through the collection and analysis of independent mutations in the absence of natural selection. Using this and similar existing data from two other bacterial species and four eukaryotic species, we find strong evidence that deletion biases resulting in repeat contraction are common in bacteria, while eukaryotic microsatellites generally experience unbiased mutation or a bias towards insertion (repeat expansion). This difference in mutational bias suggests that eukaryotic microsatellites should generally expand wherever selection does not exclude them, whereas bacterial microsatellites should be driven to extinction by mutational pressure wherever they are not maintained by selection. This is consistent with observed bacterial and eukaryotic microsatellite distributions. Hence, mutational biases that differ between eukaryotes and bacteria can account for many of the observed differences in microsatellite DNA content and distribution found in these two groups of organisms.

The problem with looking across species in any evolutionary analysis is that it’s nearly impossible to separate mutation from selection (if in “junk DNA”). To really do that, you need to look at smaller time scales.