Once they expose data structures that allow generic uses like "if size<5, add item" I'll take another look. Until then, their definition of thread-safety isn't quite the same as mine.
You can take a look at Haskell's Software Transactional Memory then. Or you can take a look at something like the Linux kernel's Read-Copy-Update (RCU) abstraction, add some persistent data structures and a retry loop on top. It's indeed a very programmer friendly way of doing concurrency.