Instead, I spent a few hours poking GTL to get teh decompression fixed and lo! it doth work now.
The error was infact pretty silly; I wasnt seeing an internal bpp after decompression. So, when you tried to get a pointer back to the data at mipmap lvls 1 and onwards you were infact getting a pointer to the base image back.. doh!
It took me far far too long to track this down, but track it down I did and its now fixed.
I'm still pondering the DDS flipping/mirroring issue. I'm thinking a normal routine should do it, it just needs a few adjustments to make sure it works on each mipmap level correctly.
I might do that now... I've got about an hour before I have to go out...