To me, millions test sounded like a good thing. The main issue I had with the description of that workflow is that if the tests take 20-30 hours to run, someone in Oracle is not spending sufficient money on build hardware! They should have tens of thousands of servers, not hundreds, and they should be kitted out with the fastest processors and NVMe drives money can buy.
If you look at how companies like NVIDIA do things, they throw ungodly amounts of compute at development and simulations. Entire data centres worth!
It's possible to have millions of tests and NOT have the problems described in that story. SQLite has a famously thorough test suite and I've seen the SQLite developers turn around a suggested improvement to their codebase - including authoring the fix, testing it and landing it, in just a few hours.
If you look at how companies like NVIDIA do things, they throw ungodly amounts of compute at development and simulations. Entire data centres worth!