Explain the anomalous one-worker ML-KEM throughput advantage

Determine why ML-KEM-768 is faster than X25519 in two exploratory one-worker PUT cells, even though key establishment constitutes too small a fraction of request time to explain the observed 4% and 6% advantages.

Background

Two exploratory one-worker PUT cells showed ML-KEM-768 throughput ratios of 1.04 and 1.06 relative to X25519. The corresponding follow-up hybrid-matrix cells did not reproduce the advantage, and the cryptographic timing contribution is too small to account for the effect. The authors identify an allocator effect as a possible explanation but explicitly state that it was not tested.

References

We have no explanation; an allocator effect of the kind described next is a candidate that we did not test.

— Benchmarking Post-Quantum Key Encapsulation in S3-Compatible Object Storage under Client Concurrency  (2610.21457 - Lingala et al., 15 Oct 2026) in Section 6, subsection “Open Anomalies,” anomaly A4