In the first case any undetected errors (highly unlikely) will have to occur in all three repo’s before you are totally out of luck.
In the second case, any undetected errors (highly unlikely) to the local backup will be propagated to the remote backups, so you could be totally out of luck.
The second case can be faster when there are substantial changes, but the first case should be faster (because of the cache) in most cases.
Personally I am using your first case.