By having 100 substeps in the model you are essentually running a 'static' crack analysis 100 times, and also having time for the mesh to be morphed in between these steps. I am not surprised you are seeing a significant time increase. On top of this, often your mesh will leave a trail of refined elements, thus each calculation takes longer than the last.
You should consider how many substeps (growth iterations) you really need/want. You can also look at the documentation on mesh coarsening as the solver proceeds to help keep the computational time down.
A further recomendation is trying to do a sub model of the portion needed for fracture.
In general crack growth analysis is a time intensive analysis, so you should also be aware and plan accordingly that these analyses will take significant time, computing resources, memory and disk space as compared to many other FEA analyses.