I probably don't have the solution you're looking for, but why not just knit your Rmd document to whatever intermediate format would get input to pandoc and just call pandoc directly for stdout capabilities?
For example, you could just knit your Rmd to md or TeX and then call pandoc directly from the command-line (or via system()). Does this get your closer to what you want?