I believe with autodiscovery this is possible.  (Vitalii will know for sure.)  Here's my reasoning:

When I started using PBS and PP, I was doing some benchmarking using ppserver.py -a and found odd results.  What happened was I submitted, say, a 5, 4, 3, 2, and 1 node job all at the same time.  What I didn't do was assign a unique port number to each job.  So, when I went to look at the results, the timings were all over the place.  The reason was I had 15 servers all looking for work.  So, it didn't matter to PP that PBS had only asked for, say, 1 node, PP saw 15 servers and used them all.

So, if you are allowed on your cluster to have ppserver.py running at all times (like, say, mpd on an MPI system), then if you start them with the -a switch, it should allow for a node to go down and re-upped with no problem--other than the loss of a node.