For instance one of the major complaints about the JNI is that adds a pretty high performance penalty to any native calls (and typically you want native calls when performance is important). Does Go's FFI not have this performance penalty? If not, why not?

Like with the JNI, Go's garbage collector needs to understand life cycle semantics of objects coming in/out of the native binding, how is this accomplished and does it do a good job? Does it complicate the garbage collection of other unrelated parts of the system?

Basically, if you are going to claim that the JNI is awful (and I think that is a valid claim) and then imply that Go's isn't, I'd like to know why that is.