I'm still following Panama's progress, but I will not start testing until it stabilizes. There are still too many API changes happening and the promised performance isn't there.
The latest bad news is that they're trying to shoehorn safety features into the memory access API. Which is hopeless without a) sacrificing performance or b) copying something like Rust (impossible). Not saying it's not a noble effort, but it will take more and more time to sort it all out. I would have preferred just being able to write C in Java (same semantics, same unsafety, same performance), then letting libraries like LWJGL handle and hide the nasty stuff.
On the performance front, there are still too many open issues with Hotspot that affect Panama performance, who knows when and if they'll be fixed.
And, as always, we're still waiting for Valhalla, which imho is a prerequisite for Panama to turn out nice. The good news is that I personally trust the engineers involved and they'll eventually figure out the best approach for both projects. However, the design and implementation issues are extremely hard and the process necessarily time-consuming.