I have some big jobs like that, although the parallel tasks can take a couple hours each. I'd start with relatively low concurrency and scale up slowly in tests. Watch your resource usage if your cluster has a limited number of resources in case cpu / memory pressure is contributing to slow-downs. Filter out most of your non-critical logs from collection. You can configure your log level in your run config:
https://docs.dagster.io/concepts/logging/python-logging#configuring-a-python-log-level-
I don't have a lot of other ideas besides that. I can imagine high parallelism with short jobs would result in a lot of dagster events flying around needing processing, so I'd imagine you'd want to allocate a decent amount of cpu for the user code and webserver components