For runs with small time steps, consider running some parts of the model less often than every time step

When running CTSM in WRF and other NWP configurations, a typical time step may be something like 3 min. It strikes me that some expensive processes, such as soil hydrology, probably don't need to be run at that high frequency. In runs with a small overall time step, would it make sense to have a separate time step for some of these slower processes?

I could imagine there being significant scientific challenges (along with some more manageable software challenges) with doing this right, but it could potentially have a large pay-off for NWP configurations.