library(rbenchmark)
library(dplyr)

Here is a bentchmark comparison between base::split and dplyr::group_split.

# Make a long table with 100 copies of mtcars
dt <- list()
for (i in 1:100) dt[[i]] <- mtcars
dt <- bind_rows(dt)

benchmark(
  "group_split" = {
    group_split(dt, cyl)
  },
  "group_nest" = {
    group_nest(dt, cyl)
  },
  "split" = {
    split(dt, dt$cyl)
  },
  "split2" = {
    dt %>%
      split(.$cyl)
  },
  replications = 1000
) %>%
  knitr::kable()

	test	replications	elapsed	relative	user.self	sys.self	user.child	sys.child
2	group_nest	1000	2.593	2.614	2.507	0.041	0	0
1	group_split	1000	0.992	1.000	0.874	0.091	0	0
3	split	1000	2.995	3.019	2.844	0.081	0	0
4	split2	1000	3.312	3.339	3.105	0.084	0	0