It could just be that the current miner implementation isn't well optimized -- it doesn't fully use all available GPU power. Of course, it's part of the design, so GPU miners won't get too far ahead of CPUs. But it gets more complicated when you compare CUDA and OpenCL miner performance.
Actually, NVIDIAs performs better than AMDs with Zcoin's algorithm.
Algorithm or its current implementation? That is my question.