Don’t know what you are talking about, but i am talking about the actual reason why replacing bufferdata is limiting the performance, and why HSA does not magically solve it at all. The pci-x bus in not the limiting factor.
You should be able to push trough 500mb per frame. It probably comes down to around 300mb in real world scenarios, but if thats your limiting factor you have bigger problems 
All i am saying is that the cpu, the driver and the actual gpu are most of the time out of sync, and operations like glBufferSubData requires at least a syncpoint on the gpu.
Even worser, i really think that glBufferSub* does also block the calling thread to shield the application from a possible Buffer.dispose()/free() right after the call to glBufferSub*. But i can test that today evening…
I am talking about the jME update loop which sequentially runs on the CPU ad then on the GPU.