I am using WGCNA to compare 2 networks of a healthy and a diseased state.  I'm using the TOM matrix as a similarity measure and creating 2 fully connected networks (healthy and diseased).  The healthy network has the highest scale-free topology using a power(beta) of 4 while the diseased network has a maximum at 3.  

Is it statistically robust to compare network properties of a network using a power of 4 and another network using a power of 3? For example, if I compare the pagerank value of node_A in the diseased network vs node_A in the healthy network.  Is that a valid comparison? The entire dataframe was normalized in the same way but half of the data was used to create the healthy network and the other half was used to create the diseased network.  I

have some really interesting findings that are a lot different than what I was expecting but I want to make sure this method is ok from a statistics standpoint. 

AFAIK, yes, the soft threshold is to find the point where the underlying scale-free topology network structure emerge. It seems that they have different scale-free topology networks, so it is reasonable to expect that they appear at different thresholds. You could check if it differs much with the same threshold for both of them.