Of course, the speed of gl4java itself doesn’t change.
But the more complex the overall engine gets, the more stuff will be necessary to do all the approp. sortings (material, spatial, …), implement the approp. culling mechanisms, implement things like transparency sorts CORRECTLY and efficiently, allow for efficient collision detection/picking, make effects not stall the pipeline, integrate the sound into the spatial structure …
There a real lot of things to do getting from a rendering API to a true 3D engine. Most of them are not that much needed my simple games, more are needed for more complex games.
For myself I can say I know quite good about 3D technology from my days when I was young, but I am very sure that the Sun guys are far ahead of myself!
No, I’m neither blind nor a purist. I just know that if I felt I’d better use naked OpenGL, I’d do it in C or C++. Just for the reason you mention: then I could use all the things around for OpenGL.
I just see it the other way round. I don’t start with wanting to use Java, then look for a 3D API. It’s more that I see the great Java3D API (together with a great network API, a good GUI API, a good memory management system) and then feel comfortable that I can use Java with it.
For me, exploiting a well-behaved, powerful system where I feel that real experts built it is just pure fun. Implementing bounds determination and frustrum culling is NO fun. Did it before. Sucked.