Investigate why multi-pack reuse is not effective

In [Feature flag] Enable multi-pack reuse of objects (gitaly#6764) we have started to roll out multi-pack reuse to our production systems. But while the feature flag is enabled for GitLab.com it doesn't seem to be effective, as clones continue to use a single packfile at most:

$ git clone https://gitlab.com/gitlab-org/git.git
Cloning into 'git'...
remote: Enumerating objects: 418862, done.
remote: Counting objects: 100% (6664/6664), done.
remote: Compressing objects: 100% (246/246), done.
remote: Total 418862 (delta 6502), reused 6518 (delta 6418), pack-reused 412198 (from 1)
Receiving objects: 100% (418862/418862), 143.36 MiB | 10.18 MiB/s, done.
Resolving deltas: 100% (323200/323200), done.

Note the "from 1", which says that we used only a single packfile.

We need to investigate why this is, as multi-pack reuse has the potential to speed up clones by a fair bit.