If the disk goes bad so as to provoke a misread of a sector, post
write, the file is effectively corrupted. If this happens with git,
the checksum calculated on write will fail to match, and the
corruption is detected.
If you're worried about that (not totally unreasonable on modern
disks) wouldn't it be better to use a filesystem like zfs that checks
for corruption at the filesystem level?
In practice, rsync works great. I pull the RFCs and I-Ds every night.