Actually, I think an intelligent cross-fade would be nice. I had thought about this before, but hadn't posted the idea yet.

The crossfade algorithm already has the sections-to-be-crossfaded in memory. It could analyze those sections and see if they are a fade out or a fade in, or if they stop suddenly or start suddenly.

For instance, the Rush song "Tom Sawer" opens with a bang that can be clearly seen as a sudden start in the waveform. You wouldn't want to fade that song in because you'd want to hear it start with that bang. But if the song before it was fading out (also clearly seen in the waveform), you could have the bang of Tom Sawyer start somewhere near the end of the fade out, but not completely after it.