I get that completely. It's just...I don't believe using 4k bit keys would expose the system anytime in my lifetime. I, obviously, could be wrong. But, I think I would take that chance. I was just hoping someone could convince me that I should't (now that I think about it).
These guys are complaining about disk space and bandwidth, not message security.
Even so, the current network can probably handle 100,000 messages a day, and the bottleneck there is some side channel timing attack mitigation code that causes the client to sleep while syncing with a peer. If you separate out the message syncing from the decryption process and eliminate the timing attack potential, the network can easily scale to a million messages a day or more.
At that point, the messages will need to be broken into 'streams' so that you can partition the traffic. The protocol supports this, but punts on the implementation details, so there's no easy way to implement multiple streams at this point in time.
But I would hardly describe that as full-on 'fail'. Everyone-shares-everything is a design feature to preserve anonymity. It's more difficult to tell who sent a message, who received it (if anyone), who was able to read it (if anyone), etc.
Reading more on this software, it seems like they try to solve the capacity/bandwidth problem by using a distributed hash table, but now the protocol requires a lot of handshaking with specific machines that has potential to remove some anonymity, and also potentially makes it easier to prevent a user from getting messages. Block enough traffic at the Great Firewall and you might not be able to get messages. [Take the above listed weaknesses with a grain of salt, I haven't done an in-depth look at the protocol.]
But in general it's probably premature to worry too much about scaling, since the bitmessage network can already handle several more orders of magnitude than the current traffic levels:
