I did have a good read of their whitepaper, there's some changes specifically related to the handling of reference counted strings for that target. The new setup is better for certain access patterns, plus having LLVM is bonus I guess, there's a lot of good work being done in terms of optimizations for the intermediate bytecode. But that's only going to be taken advantage of with new builds of the compiler. I would be an awful lot happier with the decision if there was more of a commitment to incremently update the compiler in order to properly take advantage of the work being done on LLVM, bugs fixed etc