Oh, I just re-read your title.
To clearify, you probably don't need the RefSeq information for DNA-Seq variant calling (using the pipeline that I mentioned above).
For RNA-Seq you probably will need a table of mRNA locations. However, I probably wouldn't remmend using BWA for RNA-Seq. You should use something like TopHat or STAR to handle the exon junctions.
Either way, you probably really want the coordinates rather than the mRNA sequences themselves.