Sounds strange. A test of PATN generated 300,00 by 20 dataset and 10 groups with Gower Metric worked fine. Random data willbe slow to convergence. Your 3 minutes seems too fast to me, even on a fast PC. Even after a few iterations, the groups should be ok.
It would be best to send me the dataset as csv file (you can export rom PATN if you don’t have a csv file) and zipped up. That way I can better test your problem.