(1) If this doesn't apply to your office network get a better network. Now.
Bollocks. I've had two machine failures in eight years, versus at least 20 network failures. Sure, maybe the network admins at five different companies I've worked for all just happen to be a bunch of muppets, but I highly doubt it.
That's your privilege. I don't. The only time I remember the network going down was after a complete air conditioning failure in the server room (a highly unlikely event in itself, but not IMPOSSIBLE) which forced a complete shutdown of IT services. And even then people could still work. Sure, not as well as usual but working was possible. The last time a machine failed was .. oh right. Yesterday.
Also, get better machines. The last time a machine failed at this all-macbook shop was several months ago.
In either case, you will lose unpublished work in the event of a catastrophic local drive failure.
Only in the case of a centralized system, you will also be unable to work unless the entire network path and remote server is available. This will almost never be a question of data loss but it means that you will be unable to perform version control operations until it's resolved.
The point was simply that “low” is not the same as “does not apply” and that matters when it's something which prevents someone from doing their job. Even when I worked at 100% on-site projects, I used git-svn so I could make local commits and ignore locking mishaps.
Don't get me wrong, however, I'm totally in agreement for having better tools for supporting the local, centralized workflow. The other reason I used git-svn was because merging was much more reliable and I could rebase changes to squash commits before sharing them with others. All three of those features should work well in any serious version control system regardless of whether it's centralized.