You're absolutely right: negative results require a much larger N to be compelling. Much more and better designed testing needs to be done.

The obvious effects that basic linear systems theory suggests a cable can have on the signal passing through it (attenuation, phase shifts, etc.) are all basic acoustic manipulations for which you can find well-established human hearing thresholds in academic journals. Measurable physical properties of well-made interconnects usually predict a signal degradation orders of magnitude below these human thresholds. (There are exceptions, though. I believe the MIT and Transparent cables both house passive filter networks which make noticable changes to the signal.)

The interconnect tests that I've read about in greatest detail were performed at DAL, using an N at which other subtle component differences were significant. He's rather chatty on USENET; I bet if you dropped him a note, he'd be happy to share more details with you.

Hm... I'm curious... if something like extreme #2 were in fact observed (highly significant p-value, but very small actual difference), how much would people be willing to shell out for interconnects?