It would change things primarily under the hood for table-like data structures ... imagine having a data.frame or tibble object but in memory it's stored in a very specific layout which allows it to a) make use of pre-created highly optimized library code to do complicated queries b) be interchanged with other programs with almost zero overhead, e.g. imagine doing a query on a massive database in spark and then loading it into R or pandas instantly without having to wait.

Thanks. Do you think there is a use case for traditional "small data" analysis in R, i.e. standard CSV->dplyr->output sort of transformations?

It should make things a bit faster, and a bit easier to collaborate with people using other languages.