Hi Jonne,
Phex can in fact interpret and output RSS-Feeds, they just aren't included in the subscriptions, because I was a bit vary about writing code which would interpret every xml-file. I fully trust my codings skills to write something which allows a malicious xml-file to do quite nasty things, so I kept my fingers away from making xml-readouts automatic.
But if you include magnets as <link>-tag, phex can already get the magnets from them.
Besides: With magma-subscriptions, it is "push" too: The user subscribes, and Phex gets him/her every new file automatically (I hope we'll have a file-selection dialog someday, but I gave up the try this weekend after several hours). Files which are already in the share should not be downloaded again.
I'd like it very much to have rss-feeds with magnets (though I still prefer Magma-lists, maybe because I wrote the
specs 
... I have some
neat ideas, how they could be used in the future, but I don't know if someone will take up that idea, or if I will be able to code well enough to realize it in due time).
About additional sources: I thought about http-downloads, sorry for the braintypo...
Using http-sources, you can still have the clients mainly use Gnutella, and you can use slow webspace to jumpstart the magnets. I get around 300kB/s for popular files via Gnutella without http-source, and since my line can't deliver more juice, a http-source would be mostly insignificant as soon as the download mesh starts to kick in (and if the clients are programmed correctly, they won't even touch the http-source, when they have enough altlocs and get the full speed from them).
Gnutella can definitely outperform http, at least if you don't have a highspeed webpage and serve popular files. And even if 50% of the file come from the http-server, that's 50% saved bandwidth. For a managed server you could also simply reduce the speed it gives users for the files, and Gnutella takes the lead again.