I reforwarded it through the Intel internal testers and they said this time around it was the same issue. Turns out the compiler is compiling as universal binaries then only including the PowerPC instruction set, leaving a corrupted dylib... it's a real mess on the compiler's behalf.

I have split the binaries up so the PowerPC version can run in Rosetta at quite a fast speed, but the Intel version is still being hammered out by the Intel side of the developer group (CrossWire DG/Intel)