Recent human studies particularly from the 1000 genomes project have been showing that for whole human genome, a ts/tv of around 2-2.1 is generally correct. This is only when assessing the genome as a whole.
Heng Li mentions in this thread that different specific genetic regions will display different ts/tv ratios. Looking at human exomes, it appears that the ratio increases to a ts/tv of 2.8-3.0 or higher.
There is a biological premise for this, however. In the case of exomes, the theory is that the increased presence of methylated cytosine in CpG dinucleotides in exonic regions leads to an increased ts/tv ratio. This is because methylated cytosine can very easily undergo deamination and transition to a thymine.
Interestingly, this does suggest that metrics such as GC content are linked to ts/tv. However, because ts/tv is a measure of sequence changes, it is a metric that inclusively accounts for numerous other factors, be they GC content, radiation exposure, or intra-species variation. For example, when looking at YRI versus CEU individuals in the 1000 genomes, we see that ts/tv is different between the two groups.
Ultimately, ts/tv is going to differ from species to species, and even among populations and individuals in the same species. It is therefore important to estimate accuracy another way prior to using ts/tv (for example, by SNP chip or genotyping comparison), then tweak your variant calling parameters for the highest accuracy and determine what the optimal ts/tv of your genome/exome/genetic region is. You can then use that optimal ts/tv ratio as your metric to aim for.