The performance cost of EMBM has more to do with the nature of the texture reads and there effect on the texture cache than the logic involved. On GC this is largely not an issue, since you upload large chunks of texture to the EDRAM.
I'd also not call it superior. It's a hack/approximation which can look good with the right datasets.
I'd also not call it superior. It's a hack/approximation which can look good with the right datasets.