Just to let you guys know, it is entirely possible to improve the OpenCL kernel of this miner by eliminating scratchpad registers (register spills). I am currently getting 795KH/s, a 30% increase in speed, with a RX 480 and a heavily modified kernel. It is also possible to optimize the generated GCN code and make the kernel even faster.
I am currently focused on Ethereum mining because it is more profitable, but I might release my custom NeoScrypt kernel when I have a chance.
It's definitely possible. There are 192 scratchpad regs for Tahiti. It's also important to keep the ISA size small enough to fit the GCN code caches. I have no RX 480 and cannot optimise for it right now. If you care to share your work, I'm sure some donations and thanks will go your way.