I have a new article on speeding up R by parallelization that I want to share: Speed up your R Work. I show how to speed up code from any of the packages data.table, dplyr, or rqdatatable (and it also works with base-R). In particular I think the rquery/rqdatatable packages are really going to be something big, if given the chance. Please try them out!
Have you considered setDTthreads(1) for the data.table code to avoid nested parallelism and to separate the data.table and parallel components?
Good point for timings. For actual execution I think just leaving the threads "as is" is fine.