Ouch, all night?! I'm very sorry that the pads gave you so much trouble, but that definitely doesn't sound right and contradicts my experience and the reports I've read online from others. In my Asus board, the VRMs would throttle with the stock pads, but not after replacing them with the Fujipoly Extreme Plus pads. Fujipoly pads are pretty expensive and have 2-3 times (depending on type) better thermal conductivity than any other pads on the market that I'm aware of (unless the manufacturer is insincere). I highly suspect that something else is at play here.
- Make sure that you also have the VRM controller covered (on the right edge of the top heatsink). It's easy to miss. In total, there are 21 packages under the heatsinks that need to be covered.
- Maybe the pad thickness wasn't the appropriate one (too thick or too thin).
I assume this is the voltage setting in the BIOS. If yes, then with LLC -> 6, this should be ~1.26V vcore under load @5Ghz (you can verify with CPU-Z under Windows). If that's the case, then this is a good cpu sample.
Also, seeing how many iStat readings are incorrect, you could verify clocks/temps/power using Intel's power gadget, just to make sure that the clocks stay stable under load. It also displays an accurate power consumption for the CPU (make sure you have SVID support to "enabled" in the BIOS for that).
That's not too bad and I wouldn't worry about it, although I expected a bit lower considering the low voltage/LM/IHS.
If it bothers you, you can always set the fan speed profiles from the BIOS (QFAN(F6)) to something a bit lower, by directly adjusting the speed/temp graph (Q-Fan is awesome!).
A single EPS12V 8pin has 4 +12V rails and with 18AWG wires (like most good PSUs have) you can very safely pull 7A from each rail. This means 4 x 7A x 12V = 336W from a single 8pin. A good PSU and M/B should have no problem supplying that. Even if you subtract 20-30W for the VRM switching losses, it should leave 300+W available for the CPU, which is much more than enough for any 9900K overclock (barring extreme OCing with dry ice or LN2).
The second 8pin is simply not necessary, but many manufactures often do things that might not make much sense, but they consider that will help them with sales. Like putting overclocking knobs that go to 11 (ramping vcore dangerously high) or adding VRM "heatsinks" that are more design pieces than actual finned heatsinks and on top of that, cover them with flashy plastic shrouds, reducing their cooling efficiency even further.
I'm very happy that everything turn out great for you! Like I mentioned previously, you can always try increasing the per core frequency (instead of syncing all cores to 5Ghz). That should give higher single thread performance (7000+ in geekbench). Leaving everything else as you have it now, I think that you could push 5.2-5.3Ghz for 1-2 cores without problem.
After that you can try increasing the cache frequency, again without doing any other changes. Default is 4.3Ghz, but you should be able to get it somewhere closer to 5Ghz (you might need to disable "ring down bin" first). That should give another small performance boost.