Holdon, i don't even see the need for that; if you traceroute out, the
packets will cross the exchange irregardless of whether you're announcing
it to your customers, and a message of TTL exceeded will be generated
from the exchange's address.. The important question is, should they
be globally UNIQUE for troubleshooting purposes? I think so.
When not advertising exchange point address space, the conditions with
respect to traceroute are much the same as that of RFC 1918 address
space; you can traceroute *through* it and get reports of TTL exceeded
from source addresses that you cannot reach as a destination. No big
deal to some, major diagnosis headache to others. I'd argue in favor
of reducing diagnosis headaches ... and I *hate* it when I can't
traceroute to points in the interior of providers who use RFC 1918
addresses internally. 
If the shared L2 media of an exchange point fabric did *not* have a
shared L3 address prefix as Alan Hannan noted earlier, where (as Alan
said) one party of a bi-lat coughed up a /30, then the reachability of
said prefix could be left to the parties to the bi-lat. Modern
equipment should be able to handle the quantity of secondary/alias
addresses needed to deal with number of /30s required to uniquely
number each bi-lat, no?
Stephen