At the scale of GitHub, it probably makes business sense to bucket sort time stamps instead of performing a lookup for every record for every file.
Eliminating all the potential latency due to cache invalidation probably improves user experience as well. I mean if a user experienced four minutes to retrieve dates because of network issues it would be bad even if it only happened to them only once every few months.
Also, the primary use case for git is to use the latest files and whether the latest file was from 205.6811 hours ago or 305.1342 hours ago is not a statistically meaningful use case for a tool like a webpage representing a git repository.
I mean sure it matters in edge cases for git, but a user can always look at their local repository.
My apologies to anyone who feels I am not adequately outraged, offended, or angered by vague dates. GitHub is amazing and better than any practical engineering alternative operating at a similar scale.
YMMV, good luck.