> In case 2, redundency is already handled by TCP/IP given that we
> implement the right network typology.
> 

The objects themselves can fail, too, however.  Servers are unavailable
all the time.  How does one build systems when components come and go?

Somehow, biological systems do pretty well in such circumstances.  You
can't do without an entire heart, but there's no individual cell which
is crucial.

-Lex