Looking at the spec comparison, I can't see anything that would be reason to upgrade from the D4 to the D4s.
True, but Nikon has maintained this update cycle for a long time now. Major pro gear releases every 4 years, with an update in between, which is why they call it the D4s, not the D5. So the intent is not for D4 owners to upgrade to the D4s, but rather that their flagship is at least current to within a 2 year cycle.
Now a question that goes way off topic but something that sprung to mind. How do those ultra high FPS slow motion cameras work? How do they capture enough light when they're shooting at 20,000+ FPS, which equates to a MAXIMUM shutter speed of 1/20,000? Do they just use a really high ISO?
Short answer: they use a lot of extra lighting.
A modern sensor (say, the Sony sensors used in the Nikon D800) has very low read noise (usually around 2-3 photoelectrons). The other sources of sensor noise, namely pixel response non-uniformity (PRNU) and photon shot noise dominate. Of these, photon shot noise is a physical property of light, i.e., you will only receive a finite number of photons per unit area per unit time for a given amount of incident light. The photon shot noise magnitude is directly proportional to the square root of the number of photons (Poisson distribution), so the signal-to-noise ration (SNR) can be calculated directly as P/sqrt(P).
When you bump up the ISO, you are effectively taking a small signal level (few photoelectrons) and amplifying it to the desired range. To pick some arbitrary examples, let us say you are capturing 500 photoelectrons at a given photosite on your sensor, for some given exposure setting.
Your read noise is still 3e, and your photon shot noise is about 22e, giving you a total (simplified) SNR of 500/(3+22) = 20. If you crank up the ISO, your SNR will remain at roughly 20, no matter what you do (keeping exposure fixed, of course).
Increasing the amount of available light so that you now capture 5000 photoelectrons increases your SNR to 5000/(3 + 71) = 68.
To summarise, the real obstacle here is the physics (finite number of photons, and the inherent uncertainty (=noise) in their arrival times), not really the technology. You can increase the quantum efficiency of your sensor (number of photoelectrons liberated for each incident photon), but we are already at above 50% efficiency, so I do not anticipate many surprises there. You can go wide-band, e.g., rather than capturing colour (by selectively filtering out some of the available photons), you can capture a panchromatic (e.g., grayscale) image, allowing you to use more of the available photons (which will increase your SNR).
Or you can simply use a LOT of extra lighting
