Right now ergmito_formulae, thanks to ergm::ergm.allstats, is very fast when it comes to compute simple statistics, but if the statistic is more complex like nodeicov with a continuous variable, then it can take much of the computation. Because of this, it is natural to use parallelization when we cannot recycle the statmats.