When a PSOCK cluster (maybe other cluster types too) is created by the parallel package, an Rscript process is spawned for each node. At least by default, the Rprofile.site file is read for each node that is created, which can constitute the majority of the time to create a cluster. See: http://stackoverflow.com/q/23540129/134830 Is there a way to create nodes using the --vanilla argument, so that Rprofile doesn't get read? Alternatively, is there a way to persuade R to only read the file once? It seems inefficient to do the same thing four times.