I don't think the process gap is that much of a problem, because the high-end market is mostly made up of chips like the i7-4790K.
In 2016, I'd expect Intel to stick to the same strategy: a dual/quad-core APU for >95% of the desktop market, up to $300~400, and then a different, very expensive platform with far more cores.
That quad-core design should be a very manageable target for AMD, provided that:
a) K12 is indeed a substantial improvement over Excavator,
b) GloFo's 14nm FinFET process is roughly as fast as its 32nm SOI one, or ideally faster,
b) they make a design with at least 8 cores and a good cache hierarchy.
None of the above is guaranteed, but all of it seems doable to me. It would result in a design that would be very competitive for multi-threaded workloads, and acceptable for single-threaded ones.