Now, readcsv is a whole lot faster, but writecsv is really slow. Perhaps the same kinds of issues as #3350.
Just tried a 435MB random number csv; reading it took 11 sec and writing it back out took 7 sec. Can you give a case where it is slow?
Try the file in my home directory on julia.mit.edu. Read it and then write the result back.
buffered writedlm. print substring calls write substring. fixes #3483
faster utf8 readdlm
Thanks @tanmaykm. Now, readcsv is as fast as DataFrames.readtable for UTF8 strings, and writecsv performance is significantly improved - 4x on my test file.