We were unable to reproduce the performance difference between versions that was reported.
It would be suggested to verify that nothing changed in the environment in which each version of ArcGIS Pro was running when the timings were taken. For our testing here at ESRI we made sure nothing changed on the machine the testing was performed on for each version of Pro. The input/output was coming from/to the same SSD when testing each version. No changes to OS between runs. No additional processes were running when the testing was running.
This case is sub-second and would be very susceptible to the Windows memory manager variances and HDD r/w variance, as well as being potentially impacted by any update or security software on a machine.
It may be worthwhile to see how PairwiseErase performs for this sub-second case. Testing shows PairwiseErase being slightly faster than Erase at each version.