I just started working on predicting steady-state protein levels from different codon usage bias measures. I have whole-genome sequences (it's NGS data) from different strains of S. cerevisiae and S. paradoxus and I began to wonder if it's possible that different strains of the same species could have different copy numbers of tRNA genes? If so, would this be detectable in the sequencing data (presumably the reads were mapped onto a reference genome so maybe this copy number variation would be lost)?
I'd take that computational angle and run the sequence data through tRNAscan-SE (Lowe & Eddy, Nucl Acids Res 25: 955-964). Ideally, you'd install this locally. This tool is what the UCSC folks use and it has been the best known, most widely used tRNA predictor for years. It's what we all used on Arabidopsis thaliana genome annotation back in the late 1990's.
There is also a genomic tRNA database that may have many of the predictions you seek, at least for some of your species/strains.
I would use an RNA microarray to look at those difference instead of sequencing. To delicately amplify your tRNAs in an unbiased manner would be a tricky molecular biology endeavor. I wouldn't be surprised if there were detectable and significant differences. Those experiments will be able to confirm your hypotheses regarding codon usage bias.