Serial execution, hitting PGA max per session, query slower because of TEMP spill

Small version of the benchmark database is 10 TB, "expanded" version is 100 TB (same data model and data distribution). Reporting workload run on 10 TB, then run again on 100 TB to test scalability of Oracle. One representative query runs in 10 minutes on 10 TB database, runs in 3+ hours on 100 TB database. First set of benchmarks runs all report queries as NOPARALLEL.

Of course I'd like to run these benchmark queries in parallel 8 or 16, but suddenly Oracle Database is "inferior" to XYZ Database because XYZ can use > 2 GB of the PGA equivalent for HJ without the parallel query equivalent and not have to use TEMP (and of course increase elapsed time dramatically).