I always split right before the song - after the banter, unless the banter is in some way introducing the next song.
In cases where the banter includes a song introduction, I split it right before the actual introduction. For instance, a song ends and the vocalist says "That was one from our first album, this next song is from our newest album" I will probably split it where I typed the comma. Sometimes it's easier said than done, but I listen to iTunes on shuffle play a lot, and to do it any other way results in some weird nonsequiturs.
If the talking is just unrelated banter and over about 45 seconds, I will often split it into a separate track. If the space between songs has no talking or tuning, I will leave it attached to the previous song unless it's over about 90 seconds, then it gets its own track. Personally, I do not fade my recordings in or out, and if I have to break a show into parts to burn CD's, I'll pick a spot where there's a natural pause whenever possible.
Edit- I just remembered this thread while tracking out a show, and realized that I didn't post the full story for how I deal with between-song tunings. If the band starts playing sound which would tend to indicate the next song they will be playing, I often track that with the next song. Think of Grateful Dead shows- the band would often make sounds characteristic of the next song they would play, sometimes they would change their minds, and play something else. These tunings which relate to the next song are usually best tracked with that song, in my opinion.