Gnutella.Web.Cache is a nice innovation, sense it is just a list of ip addresses it is not like you can sue it. Now to improve the gnutella network it is necessary to extend what g.web.cache does.

Along with the ip address of a user display bandwidth, how may calculations the computer can do and how many peers it can handle.
This information would be gathered by the compatible clients and sent to g.web.cache and updated when the necessary. Now when a user opens up a gnutella client It will attempt to connect to the best clients available cutting down on the time it takes to connect to gnutella, which is a common complaint. Also the g.web.cache could say what gnutella protocol it uses maybe ban .6 considering every gnutella client that is still in development uses g1. I read somewhere that one problem with gnutella is that people are still using old clients I don't know if that is true or not. I am concerned that this function could be modified and ban a client or a protocol like G2 which would be damaging.

There is a phenomenon with gnutella I believe you call it "island effect" where there may be 1 million people on gnutella you only connect to 200 or so people. This is a little more advanced and it would require a more powerful server to handle g.web.cache but it would drastically improve the gnutella network.
On top having information about each user it should also tell who is connected to who, so lets say some one with a fairly high speed DSL with a new computer logs on, the client will look at g.web.cache and make share that it will connect to at lest one another high speed user(I believe you call them ultra peers or hubs or what ever) so eventually it will link together all the high speed users and help eliminate the "island effect" in the network. Now when some dial up user logs on it will connect to what ever it can.
This is an the best architecture that I can think of for gnutella. I am share that some one can figure out a better one if g.web.cache listed what users each user is connected to.
naturally I think all the major clients should work together on this project (BearShare, LimeWire, xolox, gnucles, morpheus and shareaza)

I have seen a program that turns a computer into a "pong server" I assume it is a program that turns your computer into a ulta peer. I think this idea should be expanded. I am sure that the majority of gnutella users know nothing about pong servers which is a shame. Like if you say "I run a DC hub" or a hotline or even a eDonkey you are pretty much a p2p god. But if you say "I run a gnutella pong server" people won't know what you are talking about if they even care. I think all the major clients on there front page in big letters saying "Do you have high speed internet? If so have you thought about setting up a pong server?" The software should be compiled for Win, Mac, and as many *nix as possible. I know for a fact I can get it on a computer on a T3 line that is on 24/7 and I am sure that there is at least one person at ever univercity in the US if not most of the world that can do the same.
You should have a stelth version. Where you open it and it just uses idle resorces and no one can tell it is running there should be some command to open it and edit the options but you should be able to change the command. As well as open every time the computer is turned on. (play dumb say something like "in case you don't was to hassle with a pong server use this version")
G.web,cache should know the difference between a pong server and a regular user. Maybe the stelth version will just do calculations for g.web.chach.

I support all of p2p however as far as decentralized networks go I feel that eDonekey has much more potential.