Two things that are crucial for a wider use of R among applied researchers. The first one is data manipulation/reshaping tool. I think the package "reshape" and "reshape2" have done good job and have largely removed the barrier. The second one is a universal output processing engine, something like the "eststo, esttab, estsadd" combo in Stata. The package "xtable", "memisc", "aprstable", and "estout" have made some progress, but they all are limited to some estimation procedures and none of them provides a universal output process engine for "all" estimation procedures. One of the main reason I am still paying for Stata is the convenience of directly producing publication quality LaTeX tables from any estimation procedures without having to manually change anything.
I wish more attention can be devoted to this latter issue.