Yes. Actual benchmarking showed either no gains or performance regressions, depending on the benchmark, with occasional marginal improvements at certain array sizes due to cache hierarchies.
This is not a general "optimization" that should be done.
This is all explained in detail in multiple places linked in the article. There were multiple reasons.
1. The performance gains were unclear - some things got slower, some got faster.
2. This was deemed as a good "intro" issue, something that makes sense for a human to engage with to get them up to speed. This wasn't seen as worthy of an automated PR because the highest value would be to teach a human how to contribute.
