Of course you will have noticed the 1000pF (marking 102K) caps soldered to pin 1 and 16 of the IC. Those are other components that SNK left out of the headphone circuit on AES boards but they're supposed to be there according to the datasheet (they are part of a low-pass filter on the inputs) and are not missing from MVS boards:
A further example of the lack of attention in designing the audio circuit on this revision is the incorrect placement of two polyester caps in the L channel path:
To rectify things desolder and swap them.
How about the other revisions? Will 3-3 have the same issue as 3-4? I didn't have the chance to test one but judging by the lack of bypass caps for the audio ICs it's safe to assume it has the issue too:
I mentioned above that revision 3-6 is buzz-free. Let's take a look at the audio circuit of this board:
In this case the caps used for decoupling purposes are electrolytic instead of the usual multilayer ceramic ones. They're the 10uF caps numbered 1, 3 and 9 in the previous pics.
Revision 3-5 is almost identical to 3-6 but guess what, the bypass cap for the DAC (number 1 on 3-6) is missing:
I haven't tested a 3-5 system so I can't tell if the missing cap makes any difference.
It may be worth mentioning here that the headphone out on revision 3-6 has the R and L channels reversed. I had noticed this years ago and wasn't the only one:
http://www.neo-geo.com/forums/showthread.php?t=94456
Since then I have found why. It's another mistake of good ol' SNK designers. Apparently they realized it and made a correction on later boards by moving a couple of resistors:
