Explain the residual X-Wing slowdown after allocator control

Explain the residual throughput slowdown of the X-Wing hybrid after controlling the memory allocator in the 1 MiB one-worker PUT workload.

Background

At 1 MiB PUT with one worker, the diagnostic allocator experiment substantially reduced the mode-dependent throughput spread, indicating that allocation placement caused most of the observed AES-GCM timing variation. However, a small residual slowdown remained for X-Wing after the allocator threshold was fixed. The authors explicitly state that this residual is unresolved and that the diagnostic was exploratory.