Let me start off my saying there is a fair amount of speculation and hand-waving in what follows. If that doesn't interest you, move along -- nothing to see here! Plus, this will be moot in a few weeks when actual performance numbers start to come in.
Something jec6613 said got me to thinking. One possible explanation for the seemingly low buffer counts of the D7100 is that Nikon expects to be able to write data to the SD card much faster than in, say, the D7000. If that's the case, I wondered, just how much faster, and what effect would it have?
So, I set out to estimate the time-to-buffer-full for several possible cases. Once the buffer is full, the frame rate drops to whatever rate frames can be written to the card. Up to that point, the camera should be able to capture frames at its rated speed of 6 frames/sec.
The biggest unknown in this estimation is the speed of writing to the card. jec6613 suggests that, theoretically, it should be the rated speed of the card. I chose to examine Sandisk Pro UHS-I cards, rated at 90 MB/s write speed. But existing Expeed3 bodies that can use SD cards don't seem to achieve that, according to Rob Galbraith. He measured the speed of the D800 at 42 MB/s. (Still a lot faster than the D7000's Expeed2 system can do, and appears to be in the ballpark of Imaging Resource's tests of the D600), I would be surprised if the D7100 was slower since it's also an Expeed3 system. I'm not sure if Rob's methodology exactly translates to what I'm trying to figure out, though, so I'm going to also consider an intermediate speed of 60 MB/s. The three write speeds considered are, therefore:
42 MB/s 60 MB/s 90 MB/s
The calculation is fairly straightforward and is done in three parts:
Buffer-emptying rate (BER) in fps is calculated as card speed/file size. The file sizes I used are the ones from the D7100 spec sheet for the modes chosen.
Buffer-usage rate (BUR) in fps is the rate at which the buffer is filled minus the rate at which it is emptied (BER). Since the frame rate of the camera is taken to be 6 fps, this becomes simply: BUR = 6 - BER
Finally, the time-to-buffer-full (TBF) is calculated as buffer size (in frames) divided by the BUR. The buffer size for each mode is also taken from the D7100 spec sheet.
In these tables, I show these values for the worst-case raw scenario of lossless-compressed 14-bit (28.5-MB files, buffer = 6) and the best-case raw scenario of lossy-compressed 12-bit (20.2-MB files, buffer = 9.)
14-bit lossless (28.5 MB files, 6-frame buffer)
42/28.5 = 1.5
6-1.5 = 4.5
6/4.5 = 1.3
60/28.5 = 2.1
6-2.1 = 3.9
6/3.9 = 1.5
90/28.5 = 3.2
6-3.2 = 2.8
6/2.8 = 2.1
12-bit lossy (20.2-MB files, 9-frame buffer)
42/20.2 = 2.1
6-2.1 = 3.9
9/3.9 = 1.5
60/20.2 = 3.0
6-3.0 = 3.0
9/3.0 = 3
90/20.2 = 4.5
6-4.5 = 1.5
9/1.5 = 6
So, what does this show? One thing it shows is that the write speed may have a significant effect on buffer usage. Let's say that Nikon manages to achieve the 60-MB/s write speed. As the table shows, for 12-bit lossy compression, that would give you about a 3-second continuous burst. That compares favorably to the D300s continuous burst for 12-bit lossy, according to Imaging Resource. Obviously, we won't know the actual performance until people get D7100's into their hands and test them out. If the write speed is in the realm of what we have seen with the D800 and D600, then yes, the buffer seems on the smallish side. But if Nikon has upped the write speed, it would explain why the buffer is so seemingly small, and some of the angst about it may prove to be misplaced. We can hope, anyway.
It may be worth pointing out that along with the tradeoff between continuous burst length and raw modes (12 vs 14 bit, lossy vs lossless), another possible tradeoff is reducing the frame rate. At 5 fps, the 12-bit lossy, 42-MB/s value jumps from a 1.5-second burst to a 3.1-second burst. So if your subject can be captured nearly as well at 5 fps, that may be a tradeoff worth making. (That reasoning applies to any body to some degree.)
Finally, it's worth considering how the 1.3-crop mode performance might look. Using the same approach, I calculated burst lengths for the 42 MB/s write speed and the 7 fps of the crop mode:
14-bit lossless (18.8-MB files, buffer = 8): 1.7 s