I noticed that 95% of the time is spent in testLocalIndexScanAfterRegionSplit.
In that test we split the table 4 times, and each time wait until the parent region is remove (in a major compaction).
The test is just as valid if we split the table only twice. That effectively cuts almost 50% off the runtime. Since we test for the number of a chunks we have in the explain plan, I cannot do the same trick I did in