I suspect fearless concurrency is another motivating factor. Better perf can be achieved by squeezing more parallelism, but without borrow checking it's difficult to do fine-grained parallelism correctly.