@robsmith11
you can tune parameter c to obtain comparable sampling costs, i had direct experience with the sampler from Toni et al. 2009 implemented in ApproxBayes.jl and it suffers from degeneracy of weights, so the lesser costs in CPU time can turn out to be due to incorrect sampling
in any case
in KissABC.jl the sampling cost is tunable, a higher value of c which defaults to 0.01 will allow the algorithm to spend less time on individual samples, the quality of the posterior will not suffer a lot, but there will be some sample duplication