Here's an argument for plain linear crossfade. I get power "boosts" with cosine crossfades...
best,
J
You'll want to use linear crossfades for correlated signals (similar spectrum and same phase), cosine for non-correlated signals (noise, different frequencies, arbitrary phase, etc.). Of course for similar signals with arbitrary phase, there's no guarantee that the output will really be constant power, since you could get cancellations or reinforcements. But unless you know in advance that the phases are correlated, it's best to use a cosine crossfade. Incidentally I'm working on a frequency domain algorithm that forces mixed signals to be correlated and in phase, by summing their magnitudes. But it's no magic trick since it does this by inducing a time varying phase shift. This isn't the silver bullet that you might be hoping for, but I think it should make a neat variation of the usual chorus effect.