On the other hand, you can shutdown RUM in PG. For us, is more efficient that way. If the PG is serving multiple application endpoints (an apache with multiple sites?) you just shut down at the PG level and it's done. (It would set alarms if you have enabled the spike/drop option at the anomaly of the app) .
As @Dante P. writes, turning on RUM at the process level is much more effective and applies in all cases. If you encounter misbehaviour with RUM on, you should definitely disable it on process group level.
If you turn RUM off for one application in Dynatrace, the RUM code still runs and decides just by the application rules if to inject or not. That means for example if a user accesses the application by another URL (for example directly by the IP address) and you don't have that URL match for your disabled application, the RUMw would still inject as it is another application.