One change of code and a recompile later and omfg, it works!
This'll teach me not to read the docs properly in future, which I need todo.
Currently its a mess which is copied, I need to fix my DDS reader so that it properly loads the 16bbp uncompressed DDS into ram then I can do some really testing with it all.
I did wonder if a DDS compressed => uncompressed routine on the 2nd CPU might be worth the effort, but then I realised that I'd either have to pre-uncompress the data as it was required by the blitter or go fully software on the drawing. So probably not worth the hastle given the textures will be small anyways.