Highly likely as the 'anti-parallelism' stuff in the hefty part is far from sufficient. Too little branching and not enough random memory access (which gpu's are very bad at), let alone plain memory usage. All other methods are not sufficient to stop parallelism as the bottleneck is memory access time/bandwidth.

All the other algorithms can (except for Groestl) be easilly put in a opencl/cuda file and run. Some tweaking and we might see some miners in the next weeks (as there are at least two running now).