Using @sipa’s function, I have found though simulations that the smallest change created is much larger than what simple SRD produces. However the problem with a higher mean number of UTXOs persists and is slightly worse with this modification. But fees are still the same, and we are unlikely to be grinding things into dust. The question then is are we okay with having more UTXOs or should we attempt to consolidate as much as currently done.
Part of the difference is due to the fact that currently we spend dust outputs, but simulations show that even this does not entirely account for the higher number of UTXOs.