We already know that a CPU miner can in fact do 10x better than Zcashd - this is almost the case for very slightly tweaked reference implementation of Equihash, as I mentioned in the other thread here: Breaking equihash in Solutions per GB second - #42 by tromp
It is not unexpected that someone actually optimizing this obtained further speedup.
I would support a delay in the launch, but (luckily) I have no say in this, and I suppose there are lots of reasons why Zcash wants to launch on schedule anyway. Hopefully, someone will not care about a potential temporary gold mine so much and will open source an optimized implementation before launch.