Well I had the following idea but my FTP was too slow for it and I didn't find a well working FTP to Drive Letter mapping tool.
The idea was to create a fast online resource for soundbanks which contains all the soundbanks from wsbdb.org (unzipped, of course).

1. Remove the "Speech" folder in the W:A's User directory.
2. Map a fast FTP server to a drive letter. Actually, \\BLA UNC links also work, but I didn't know how to create a resource in the web which can be accessed like that (maybe only with VPN).
3. Create a symbolic link to the speech directory on the mapped FTP (or whatever resource), like
mklink /D Speech "Y:\files\worms\speech"
4. Let W:A read the stuff from the ftp since it thinks the symbolic link is the real directory.

Sadly my FTP is horribly slow (including the awful "NetDrive" tool to map it to a drive letter which even crashed several times) and the game fell back to the default voice after some time, but I was able to hear the samples in the team editor in the frontend (and every sound needed to load like a second). I'm not quite sure if someone can get this working more fluently with a faster server. Any ideas or people who wanna try this?

I know a buddy which is very open in terms of FTP. I might ask him on this

It's an interesting idea, but the trick is making it work. A lot of the humour from the soundbanks comes from the timing, and an FTP call, download and play is going to take, at best, a second or two. That's going to bugger up the timing and ruin the flow.

The alternative, perhaps, is to have the game quietly downloading individual files as gameplay progresses, but even then if it goes, say, alphabetically it can't call JustYouWait.wav if it hasn't downloaded it yet. So the trick then becomes predicting which soundbank files are likely to be needed first - YesSir.wav, Hello.wav, ComeOnThen.wav, Fire.wav, Die.wav and WatchThis.wav would probably be good starting points.

This is to say nothing of the fact that if this ultimately works and is added to the game by CyberShadow and Deadcode, Team17 (or whomever ends up hosting the soundbanks) wouldn't legally be able to host soundbanks based on copyrighted material. This means you still wouldn't be able to share that really really really awesome Dragonball Z Abridged soundbank you're using for your team.

Yeah, I actually had to wait 5 seconds for a YesSir.wav to come up, the game froze in that time of course. However I bet 3 seconds of that came from my uber slow FTP and this dumb NetDrive program which doesn't work flawlessy.

What you really need is a wormkit module that creates a temporary directory, downloads the soundbank to that directory as soon as it is known to your computer, perhaps compressed even to make it faster, redirects FS calls to there, then deletes it afterwards.

To let it deal as a cache I think it shouldn't delete those.
Well, if you ever get the chance to have played with ALL the soundbanks from wsbdb.org, you have "only" 1.7 GB, but you will never play with ALL of them :D

Sadly I'm not good enough in C++ to try something like that, but if someone could call a C# program or pass it a message from a C++ wk-Module for me, I could do that :X

gotta admit... these speechbanks kinda suck anyway.