By popular demand I have some experience teaching R to novices. Being a mathematician with a fair amount of computer science experience, I always found myself rather uncomfortable teaching these R workshops and courses. In particular, students that know how to program in e.g. C++, C# or Java, find themselves totally lost by its peculiarities (or maybe even misconceptions???).
John really nails the issues in this talk. Honestly, I was somewhat sceptical seeing a talk on R in the Lang.Next conference, but I totally subscribe to John's point of view here. Statisticians love R, I do not... In particular lack of decent tools is unbearable once you are used to IDEs like Visual Studio 2010. Indeed there is quite some room for designing a suitable and reliable DSL for statistics and the proper tools that go with it! Maybe call it Q#++?