microbenchmarking of dplyr and base R