The R programming language has gained huge popularity among quantitative analysts in business, industry and academia.  Many of these users have exceptionally large computational needs, so various forms of parallel R have been developed, including my own package, Rdsm.  In this talk I will evaluate the range of applicability and limitations of the various approaches to parallelize R, using examples such as All Possible Regressions and clustering algorithms.  I will argue that certain properties of R make GPU the most promising in many (though not all) classes of applications.  Although this will largely be a software talk, I will also discuss several mathematical/theoretical issues that affect R's ability to become truly parallel.