I just figured this out, probably a standard trick that's been around forever but thought I'd share it anyway...
Say you've got two vocal harmony tracks, great takes, in tune etc. Except, one of the tracks ends just a little later than the other one. On playback, this sounds like a mistake, someone sang longer than they should have.
The fix I found is to side chain a gate on the track that sings too long.
* Insert a gate on the track that sings too long (I used a FabFilter Pro-G).
* Create a send on the other channel and point it at the side chain input on the gate.
* Make sure the side chain is listening to the other channel (Ext on Pro-G).
* Fiddle with the threshold and ratio controls on the gate until you get the cutoff characteristics you're looking for.
Depending on what you're going for you might want to make the gate tight or a bit sloppy, production choice.
Using this technique you can get tight vocal harmonies without losing the human bit.

My solutions.
 
1:  retrack it if possible to fix.... punch in works just as good. having singers who know the song phrasing helps too.
 
2: if #1 is not an option..... use envelopes to shut one or both of them off at the appropriate time to avoid the out of sync sound from the vocals not being together in their phrasing.
 
 
I have had vocal harmony tracks come in from other studios for a project I was mixing and this very thing you mentioned was evident.... the "S" sound at the ends of words is very apparent when it not in synch due to the singer not paying attention to phrasing.  Also late or early starts will cause the same headaches for the mix engineer.
 
Insert envelopes in BOTH... or ALL vocal tracks and listen to that specific part.... zoom in and manually add nodes to lower the volume or mute the one that sounds off the most. Often, the listener is not even aware that the very last few milliseconds of vocal track 3 was pulled down and out of the mix, then brought right back up again for the next line.
 
This song: http://soundclick.com/share.cfm?id=11962059  has IIRC a total of about 10 vocal tracks between the leads and BGV's.  There were several places where I had to deal with the harmony endings not ending together.   Some places I simply pulled the entire section out.... it became a creative decision to get this mixed. It was all done with envelopes. There is at least one still in there that I missed before the final mix....

anyway... that's how I do mine. It seems much simpler than side chaining a gate, creating and using a send and then fiddling with the settings.

If I had done it with automation envelopes, I would have been at it all night tweaking and refining them and it probably still wouldn't have been right. Side chaining the gate and tweaking the settings took about 5 minutes, about 1/2 that time was learning curve on the gate (I just got it), and it's pretty close to perfect.

This is a great historic tips thread.  We have the evolution of vocal editing all in 3 posts.
 
Obviously the original method is to re-record and get it it right..
Then there's automation with side chaining gates.
Once DAWS become available we can automate envelopes to force the timing.
 
 and the "modern" way...   Melodine, so everyone is singing in perfect time and pitch...

TomHelvey
If I had done it with automation envelopes, I would have been at it all night tweaking and refining them and it probably still wouldn't have been right. Side chaining the gate and tweaking the settings took about 5 minutes, about 1/2 that time was learning curve on the gate (I just got it), and it's pretty close to perfect.

 
Can you set up side chaining on 10 tracks to get them all automated to do what needs to be done correctly in 5 minutes? I would thing that would be a nightmare. Just asking.... I can see how it would be easy to use on 2 tracks and set up correctly..... but 10?
 
I'm guessing that I probably spent 30 minutes to maybe an hour on the 10 vocal tracks since most of the singing was in sync....and most of the endings and beginnings of the phrases were also correct..... it was not a major job to manually fix the places that were out.  The biggest part was identifying the places that needed work and then finding the track or tracks that were out..... and then deciding which ones to leave as the main one and which ones to edit.  As far as spending all night..... no, probably not. You see, all I did there was to pull the track that was out of sync down.... basically a muting function. And if that didn't sound good, or was unbalanced, I pulled both or all the BGV harmony tracks down as well..... better to have NO BGV than to have something that didn't sound right or balanced. The down and dirty part was quick. Then as I would listen as I worked on other aspects of the mix, I would hear something and a quick edit fixed the nits......
 
I guess it's all in how you want to do things because I have learned over the years that there are no "right way" or "wrong way" to do something and there's always generally more than one way to do a particular thing.... it's all in what makes you comfortable and gets the results you're looking for. And to me, that's all that matters.
 
As Karyn pointed out,,,,melodyne would also be capable of fixing this..... but timing issues are not automated in Melodyne like pitch issues can be. But, even there, I would never use the automated pitch correction. Manually is always more precise. And yeah, ME can edit the timing with great precision and results.
 
 
 
 
 
 

My goto tool for this type of correction has always been V-Vocal.
 
It takes seconds to nudge an errant timing into place (I intend to fully explore Melodyne in the near future, but currently, that's not an option for me)

take the track that's too long and export it, import it in to Soundforge, reduce the time "slightly", re-import it, blend it in with the original where the original fades out at the appropriate time, and the shorter (slightly) track fades in? It's a PITA, but if done correctly, it's smoother than trying to use gates or envelopes. This only works with minor changes, bigger changes must be re-tracked. 
 
H-

And you can also stretch/compress the relevant words/audio in Sonar and then bounce to clips. this way you get the ending consonants (or vowels) still intact and is about as quick as identifying the problem clips.
 
 
 
 
Edited for clarity