TDIF interface, pin-out: 

              ____
             |    \__
                      \ _
in 1   -------- 13        \
                     25 ---------- gnd
in 2   -------- 12
                     24 ---------- gnd
in 3   -------- 11
                     23 ---------- gnd
in 4   -------- 10
                     22 ---------- gnd
in LACK ------- 9
in EMPH ------------ 21
in fs0 -------- 8
in fs1 ------------- 20
                7 ---------------- gnd
out fs1 ------------ 19
out fs0 ------- 6
out EMPH ----------- 18
out LACK ------ 5
                     17 ---------- gnd
out 4   ------- 4
                     16 ---------- gnd
out 3   ------- 3
                     15 ---------- gnd
out 2   ------- 2
                     14 ---------- gnd
out 1   ------- 1     ____
                   __/
             |____/
Lots of ground! I suspect you would be well advised to carry these ground connections in parallel with the digital signals throughout the length of the cable.

The connotation of the 'DATA' signals ( {out , in} {1 , 2 , 3 , 4) ) appear throughout the design of the whole machine. At all other places, these four signals seem to correspond with the one track pair each: DATA1 is channel 1/2, DATA2 is channel 3/4, and so on. Draw what conclusions you wish from this. I suspect that the TDIF DATA signals (in and out) indeed carry one track pair each, alas, I don't have a logic analyzer to verify this. If that is the case, it could make it possible to build a completely passive digital patch bay where you can patch the signal in pairs, possibly even within the same machine. But don't come to me and complain if you try this and it doesn't work!!

UPDATE! I've heard reports that this indeed does work. So it's entirely feasible to build a digital patch bay as described above. A friend of mine has already made one. Have fun!

Also note that on the drawing I have, "out fs0" was mislabeled as "out fs2". Further inward in the drawing the signal is refered to as "fs0" though. (I guess the LSB was correct..)