Live person interacting with a puppet

I think that timing is work pretty well... If you used the other person's voice for the puppet you could just both talk and record both voices in one pass and feed that into CH (then delete the visemes for when you are talking). But as soon as the other voice is you, its just going to be hard to get timing right.

You could just read the other character's lines in your head so its just you talking then reading the next line in your head (nobody else reading the other lines) maybe?

Another trick where you do the restarts (I assume a mistake was made so you are editing and trimming) is you can hide them a bit by using Prem Pro or After Effects to cut to a zoom in on yourself or the puppet. The can scale the video in, using keyframes or similar. Then the splice looks like it is there for a reason (the zoom) and it visually looks a bit more interesting rather than the same camera shot all the way through.

But I think you have done a good job already. If you are doing two voices, getting them synchronized is essential.

Oh, there is another trick I have used sometimes - in Prem Pro you can splice the video when the puppet starts and finishes talking, then change the speed of playback for just that part. E.g. if you paused for 2 seconds but it should have been 3 seconds, you can change the playback rate for those 2 seconds so it slows down for a moment then speeds back up again. It can look smoother than a video cut.

But I think what you have is good. It will just take time.

I think you are on the right track.  

Here's an idea you might try:

For this type of scene I'd suggest you have the person doing the voice for Mario just stand off screen to the right.  They would then do the Mario lines conversationally with you.  This would have a few advantages over what you are doing now:

1) You would have someone to look at and interact with, which will help you do a better performance.  If your Mario performer stands in the right location offscreen (you will want to experiment with this) it could really give the impression you are talking to a puppet when you add Mario in later on.

2) The timing of the lines will be a lot more natural (because you are having an actual conversation.)

Once you have your audio totally recorded, you do a compute lip sync from scene audio for the entire scene. 

That will give you visemes for both you and Mario.  Now just go into the timeline panel and select the visemes you don't want (the ones for you) and delete them.