The current blob repacking code basically works like this:

• find all unused blobs
• find the pack files for these blobs
• copy all blobs in those pack files into new pack files
• update the index
• remove the (now unused) pack files and the old index files

This is not the most efficient way to do it: If just one single small blob in a large pack file is unused, the whole pack file is rewritten. Optimize the algorithm so that packs are only repacked when e.g. at least 10 or 20% of the pack's size is occupied by unused blobs.