Currently I work as engineer in video games, and I would like to say that next step is just to write software completely without malloc+free and GC : you just allocate virtual pages directly from OS, create your own pool allocators in them, and that it. When you need to "free" stuff you just wipe whole virtual page, O(1) free, faster then anything you could imagine. We have only three special cases : string operations (usually implemented with special string pool), heap (malloc + free) for libraries that cannot live without it, and everything else on stack.

That said, I honestly don't understand why everyone only talks about GC and malloc+free abstractions, because they simply do not materialize in physical world at all (even if malloc needs a new page, it will call kernel for it), cpu/kernel do not have such mechanisms, so this is only abstract limitation in developers heads. So why not just talk (if we actually end-up in situation when we need to talk about memory, 'cause usually it's just work) and measure about what's actually real?