R as a language could use some help. It is painfully slow, a big memory hog (since it copies large objects with abandon), and has lots of language "gotchas". We had a saying where I work: "R is really fast if you write it in C".

However, the libraries are great. Anything you'd want in statistics is already there. So I do use it all the time.

Just to say something nice, I do like data frames (a two dimensional matrix, where each column can have a different type).