Briefly discussed here, ATTOW, the performance of merging spills is substantially slower in the long running sorter tests, to the point where the number of keys involved in the test was cut from 800K to 100K (note that we still have basically the same coverage at this new lower level).

Primary driven index builds are actively being optimized and this disparity may shrink or go away entirely. This ticket simply tracks that there is one and can govern any work specifically done to address it.