I wouldnt think arrow would be a significant help to you.
I'm guessing that the slowest part of getting the csv's in is because its not a data efficient format, maybe have your users zip their csv's for smaller file transfer times, and have you code unzip them in a local tempdir. Then use data.table fread to read the csv.
If the users are r-users and prepare the data as dataframes before exporting to csv, have them export them with library(fst) , and evolve your app to know how to read in fst files. my 2cents