Why don't they add another 4 cores of CPU@2.1 GHz? At 14nm process the power and die size wouldn't be huge. Is the leaked number final spec or not?
Well, like shifty already worte, we don't know.
But, what is done with the CPU in current games?
in the Old generation, the PS3 had a 1 core CPU with 7 SPEs. The xbox 360 had 3 CPU cores with HT. so multip-platform titels most times only used one "real" cpu core and tried to use the SPEs. As we know, the SPEs are not really good for general CPU workloads, so most times they were used to help the GPU.
Even on PC most of the time the CPU is doing work for the GPU (e.g. draw calls). That's what really can change with Mantle/DX12/Vulkan. So in future (and some current titles) the CPU has to do less and can do other tasks. This might not count for consoles, but as most multip-plat titles get also released on the PC you really have limited CPU time for non-graphical tasks in <DX12/Vulkan.
So, where are we now? In theory we should now have more CPU-time to use for AI, physics etc. The problem is (at least on PCs) you don't know what CPU your customer has. So you can't assume the customer has the latest/best CPU. Game mechanics shouldn't change with CPU speed, so you are again limited to use now "free" cpu resources to tasks, that don't influence the gameplay (e.g. physics or again help with some graphics tasks). It is easy to scale the graphics part of a game, so the game can run on a low-end GPU and on a high-end GPU and the gameplay doesn't change.
With the PS4k it is the same thing. You must provide a game, that runs on PS4 and PS4k, so in most cases, you just can't saturate additions CPU cores because the game shouldn't change. You've got a little bit more horsepower, so you can do the caculations a bit faster, that's all. This might be needed to saturate the GPU workloads, because the GPU is a bit bigger and you still need to provide some data for it, so you need a cpu that is at least a bit faster.
You can't reduce AI, because if you do the gameplay might be broken. Scaling graphics is also the easiest part, because the user might be able to really see the difference, so investing here might be worth the development time to do so. Less-visible tasks don't pay off that much. E.g. audio calculations (which are done by the cpu). They get better, but we already reached a level of audio quality, the "normal" user doesn't hear a difference, so why invest more development-time here?
One example that came to my mind, even the Wii was able to play "Call of Duty - Black Ops", so you might imagine, how much "CPU" time is really needed if you just throw away some graphical stuff.