There are three things that explain the differences:
1) The main difference comes from the clustering method, with the current default using fraction of native contacts (FCC) while the example was run with RMSD clustering. FCC gives a more fine grained clustering
2) Running on different hardware will also cause some differences. You run might have gone to the grid.
3) Such computations are chaotic in nature. So small differences can be expected (also linked to 2)
I rerun the demo on our own cluster, both with FCC and RMSD clustering. Here are the corresponding plots.