I'm using Git 1.7.1 configured with git-http-backend (Smart HTTP). I wanted to see how much faster this was compared to the git:// protocol, since now they are supposed to be on par.

Turns out git:// protocol is compressing more objects (33,229 vs 21,193) for the exact same repo (even verified client side), and thus creating a packfile that's about 30MB bigger for my project.

Is there some reason why git:// protocol needs more objects?

git:// Output

remote: Counting objects: 44510, done.
remote: Compressing objects: 100% (33229/33229), done.
Receiving objects: 100% (44510/44510), 124.07 MiB | 2.84 MiB/s, done.
remote: Total 44510 (delta 22755), reused 15866 (delta 7516)
Resolving deltas: 100% (22755/22755), done.

http:// Output

remote: Counting objects: 24299, done.
remote: Compressing objects: 100% (21931/21931), done.
remote: Total 24299 (delta 7517), reused 0 (delta 0)
Receiving objects: 100% (24299/24299), 95.95 MiB | 2.41 MiB/s, done.
Resolving deltas: 100% (7517/7517), done.

May be: via git you download files from all branches and you can quickly switch between them, via http you download only master branch.