Why is the first deserialization slow in the container?

The serialized function is complex， which introduce some dependencies， but the number of lines of code for the overall function is not long.
Running outside the container is as fast as a non-first run inside the container， What should I do to speed up the first deserialization in a container？

The issue names no source file or test. Start by reproducing the first-versus-subsequent deserialization timings inside and outside the container, then trace the serialized function and its dependencies to isolate the first-run cost. Done means the first deserialization in the container is measurably faster, with the comparison documented.

Written by the indexing model from the issue text.