Thankyou for the first constructive workable suggestion had so far. However,
this has two problems.
a) RIPE fidn't give me the first /19 in a shorted prefix block
( its x.x.160.x and .192.x is used), but no matter, I'll renumber
if necessary 
or persuade them to give me a /18 as well so I
can do the above (hopefully).
b) The /21 advert may be inbound filtered by a.n.other, which will be
fine if it has an AS-Path through me (as the less specific route
will work the same way) but won't when that path goes through the
other provider with whom they are multi-homed, as the /21 will disappear
entirely (3rd parties, i.e. a.n.other's customers will see neither),
the /19 will be the only thing that is visible, and I'll just black
hole their packets.
Anyway this is several times better than the swamp. Oh well.
Alex Bligh
Xara Networks
People on this list of lists seem to be focussing on squeezing more
aggregation out of assignment and routing policy's.
But that won't work. The Internet isn't organized in a geographical way nor
in a very hierarchical way.
There are less than 2000 Autonomous Systems, but more than 30000 routes.
That means an average of 15 routes per AS. That's the real problem. And
we know how to fix it too: renumber.
But whatever happens, the routingtable will continue to grow. Get used to
it. At present I can still run full routing with only a Cisco 2514 so the
problem is hardly as big as some people (Sprint...) like to tell themselves
and the rest of us.
We pay those backbone people big money, so either they do their job and
quit complaining about their overloaded CPU's, tight memory and other stuff
or they find themselves another line of work.
I'm getting sick and tired of reading how everybody will bend over backwards
to find a way to live with an utterly ridiculous policy of some dinosaur
company with a flawed network because they pay their lawyers more than their
engineers.
Maybe they could get away with that in the good old phonebusiness, but not
on the Internet.
Iljitsch