What I think most people are missing in this argument is that the whole discussion might very well be mooted by a huge factor in tRNA evolution that no one seems to be taking into account. I'm talking about the fact that tRNAs are insertion targets for various kinds of mobile genetic elements, from phages to plasmid-borne genomic islands to transposons. As Jörg Hacker and Elisabeth Carniel point out in "Ecological fitness, genomic islands and bacterial pathogenicity" (EMBO Reports, 2001):
Genomic islands are part of the flexible bacterial gene pool and are somewhere between 10 and 100 kilobases (kb) in length. They frequently harbor phage- and/or plasmid-derived sequences, including transfer genes or integrases and IS elements. These particular blocks of DNA are most often inserted into tRNA genes and may be unstable.(Emphasis added.) Transfer RNAs are constantly being "inserted into" (and next to, not always into) by mobile elements, a phenomenon that's been well studied not only in bacteria but in yeast and elsewhere. Over evolutionary timespans, tRNA genes are duplicated, then disrupted, over and over again, by mobile DNA elements. These elements (whether from phages, viruses, transposons, or what have you) are known to have played (and continue to play) a significant role in shaping genome diversity, across all taxa. This is not a trivial factor, in other words. Transfer RNA genes are insertion hotspots. Surely the patterns of tRNA disruption caused by gene-hopping, over the eons, cannot be unimportant in the determination of codon usage patterns.
Many examples can be found of ancient tRNA signatures inside the tail ends of protein genes, no doubt leftovers from millions of years of insertion events.