I stand by my statement for the GTX680 6+8. The reference design GTX 680 is a 6+6 pin card, and actual power measurements revealed by sensors in the Mac Pro prove that sticking an 8-pin connector on there doesn't change the power draw or the performance one bit, whether you supply 6 or 8 pin power to it. Those two extra pins on the connector are about as useful as your appendix.
Just to clarify: the ONLY purpose of the two extra pins on the 8 pin connector are to indicate that the board may safely draw up to 150 watts on the connector. The connectors are specifically keyed to be compatible, the actual power lines between them are identical, and the cables themselves are capable of far larger loads. The board designer can choose whether or not to respond to this information, and this will vary from design to design. It is incorrect to draw a broad conclusion based on the single sample of one design.
And to be clear: I am not arguing, nor trying to prove myself correct. A combination is either within specifications or it is not. Many, many systems operate outside of specification. Whatever choices anyone makes given this information are entirely their own responsibility.
[Edit/addition]
In the specific case of the GTX 770, the overall power specification of 230 watts is very slightly above the maximum specified of 225 watts. If there were a way to guarantee that power was drawn equally from the slot and the two connectors there would be a very, very slight probability of failure. Without analyzing the circuit, there is no way to know where the card will actually draw its power - the card's design could legitimately draw 150 watts from the 8 pin connector, although that is unlikely. It is also unlikely that the power draw is equal across both connectors and the slot...
[second edit/addition]
It is clear reading the other threads on this topic that there are some basic misunderstandings of circuit design. The multiple power sources are not a "pool" to draw against. People are surprised at the uneven power draw from the various sources when they bother to measure... this is by design!!! For practically all consumer designs, each component on the board that requires power will be assigned to a single power source - this will not change over time. Components will NOT draw from multiple sources; think of it as three separate power circuits on the board. This is why it means nothing to say that the reference design uses two 6 pin connectors, so the non-overclocked boards that use 6+8 pin connectors should work. Unless these cards were laid out with the same chips assigned to the same connectors, they could have vastly different distributions of power. The board's designer is within specification to put whatever combination of chips he chooses on the same power connector as long as he does not exceed it's specification. So a non-overclocked non reference version of a 225 watt card in the 6+8 config could draw 145 watts on the 8 pin, 5 watts on the 6 pin, and 75 watts on the slot pins if the designer wanted to. And it is also clear that in the other threads that have actual measurements, people have observed in excess of 125watts drawn on the 8 pin connector - this is far beyond the "headroom" of a typical design for a 75 watt circuit. And to be explicit: I have no idea what the actual headroom is for the Apple motherboard. This may or may not be a serious issue.
As to the "it works on my machine" comments - I do not dispute this - I believe you. When designing for mass production, desired failure rates are on the order of 1 per multiple thousands of units (depending on the volume, this can be 100s of thousands). As long as you don't go hog wild, the odds are greatly in your favor - but SOME machines will fail.
Perhaps another analogy - your car's engine has a redline. The engine will not blow up if you exceed the redline slightly occasionally. Some engines can exceed the redline regularly and for sustained periods. And like the circuits, the probability of failure is related to how far you exceed the limit, and for how long, and in what environment, and with what cooling. Would you think it good advise to tell people to ignore the redline because you exceeded it and your engine didn't blow up? Just like the redline the power specification is there for a reason - reasonable statistical confidence levels of longevity over sample pools of multiple thousand units.
Personally I don't care one way or another about what someone does with their own machine... I'm only hoping to help people make informed choices.
-JimJ