i think you misunderstand. rob pike wasn't talking about anything
as specific as a pots interface.
the design of plan 9 allows one to put together a (up-to) per-process
uniform namespace that looks largely like a traditional unix name space.
but it is in fact radically different under the covers. parts of that namespace
can be on other machines. you might have a ftp file server serving /n/ftp/...
locally and a remote plan 9 machine serving a big archive mounted on
/n/archive. you might have a the kernel cons file server serving /dev/cons
and other files. you might have /net.alt (and ip stack running on exotic
networking) imported from another machine. as you can see the naming
is uniform (i'm glossing over a bit), and we can mix protocols at any level
and replace any level with mount or bind. this is all because of uniform
naming. unfortunately as rob himself pointed out in "the hideous name"
naming schemes like uris really mess this up. and of course the reality
of other systems does too. so one ends up stitching together a frankenstein.
please check out the papers. my ramblings can't do the system justice.