Why is Hyperlane's QPS so low in the plaintext benchmark?

[ltpp-universe]

I saw that in the [TechEmpower benchmarks](https://www.techempower.com/benchmarks/#section=test&runid=8577540e-9440-4968-9fa9-b3e12ead02af&test=plaintext), the hyperlane framework performs very poorly in the plaintext test scenario, with only 1,090 QPS. Theoretically, it should be higher or at least similar to the JSON scenario (which is around 900,000 QPS). I want to understand why this is happening.