As long as you aren't native, that is.
Although that's unrelated to the discussion at large.
Is there a C programmer on the planet that does production builds with a non-optimizing compiler? How is your comparison even remotely relevant to the GP?
The advantage of going the native route is that it gives you improved observability and post-mortem. The JVM is a VM (Java's) inside a VM (the OS's), and that additional layer doesn't come for free.
Some of those are actually LLVM frontends.
As for the type of hypervisors you're referring to, those things are effectively half-baked operating systems. That's why something like KVM is better -- you're not pretending to be a non-OS while slowly reinventing a wheel that the *nixes have and always will do better.
I think we're beginning to wander massively afield though! :)
To my mind, the idea of porting is that you face a choice: you can recreate the JVM functionality in your own runtime, or you can port to the JVM. Porting to the JVM is certainly non-trivial, but you then get thousands of man-years of work in return. It's certainly worth doing a project to see whether the porting effort pays off; especially if this done concurrently with working on the language's 'native' runtime.