Perhaps the R version of the microbenchmarks could be run on Revolution R Open, http://mran.revolutionanalytics.com/download/. The major difference between RRO and other versions is that RRO incorporates MKL. For one or two of the benchmarks this could make a difference.
By the way, I was also going to suggest applying the byte compiler to the functions being timed but I see that this has already been done.
