I thought dplyr (note, not plyr, dplyr is the next iteration of plyr implemented mostly in native code) was pretty much as fast as data.table.