The problems isn't the software - they physics of the thing is really simple - the variable factor is how it'd been recorded. As said, the cancellation is simply done by looking for things that are equal in level in both channels. This is normally the vocals (unless it's a duet). One channel has it's polarity reversed, then when the two are brought together to make a mono track, the common elements vanish. The problem is, most recordings have stereo reverb and effects on the central voice, so those don't go, leaving you with a phantom reverby mess where the voice was - sounds rather ghost like. You'll also lose any other instrument in the middle - so you may find the bass vanishes, as may some of the drums.
So audacity, adobe audition, sony soundforge, steinberg wavelab all can do it - but how well depends on the material, not the software.
The simple answer is that none of them do it well, but luck sometimes works in your favour!
Paul