Maybe it's the Language Barrier
Aswath takes a shot at some claims in the CommsDesign interview of Niklas Zennstrom. Reading the interview, one has to wonder: is Mr. Zennstrom intentionally being misleading about how Skype compares to other VoIP architectures, or is he just uninformed?
In answer to a question about how Skype claims better quality of service than other VoIP providers and regular phone lines, Mr. Zennstrom replies that Skype's peer-to-peer technology means that "the connections between end users are set up directly over the Internet to end users, rather than having to go through a central server somewhere."
But in every VoIP architecture, the (media) connections between end users are set up directly over an IP network (the Internet or a private IP network). The only case where the media connections do not go directly between the end users is when a Session Border Controller is in the media path - but this an exception in the same way that Skype uses supernodes for NAT/firewall traversal, in which case the Skype media connection does not go directly between the end users either.
The call/session control information will "go through a central server" in an H.323 or MGCP network (though it may not in a SIP network); however, how call/session control information is handled has little to do with sound quality.
So is Mr. Zennstrom intentionally mixing call/session control information with media connections, or does he not understand the difference?