Clustering is not in our application as such.
A non-optimal but possibly good workaround may be:
1) manually distribute the 500 addresses into logical sets, e.g. by street or district;
2) find the best route for each set and note the time;
3) combine sets into clusters of one hour.