We may want mechanisms where we downgrade fetches to blob-less and tree-less fetches when limits are hit. This would drastically decrease the amount of data fetched in some larger repository cases.

However, we may want to think COB data, and still being able to fetch those references so that we can still browse issues and patches.