thanks again for your valuable time, I really appreciate it.
I'll go ahead and start experimenting with lower values of PMBPS by setting it manually as described in ID 1269321.1
Initially I thought PMBPS has to do something with parallel ("Parallel MB/s" or something) but it seems like this is the "maximum megabytes per second of large I/O requests that can be sustained by a single process" (from ORACLE Docs).
I think the calculation of this is simply wrong. The value of ~1GB/s can be reached (in our case) with maybe 16 or 32 processes accessing the file system at the same time, but in no way by a single process. Maybe ORACLE mixed up something in their I/O calibration package. I also noticed that the bug "Bug 10180307 - Dbrm dbms_resouce_manager.calibrate_io reports very high values for max_pmbps [ID 10180307.8]" is already fixed in 184.108.40.206 Patch 11 or 220.127.116.11 onwards. Installing one of those might be an option once I can find time for the next maintenance window of this database. Untfortunately I can not shut it down at the moment and try.
Thank you as well for the explanations on how to set the degree of parallelism manually. I marked this question as "answered" since I feel we discovered the root cause of ORACLE not using parallel query in my case.