This work was previously reported on Memestreams back in November 2005, when Markoff wrote about in NYT. There must be some reason why it is being revisited now, but it may not be publicly obvious. One presumes that it came up during the Here's Looking at You... session at Phreaknic. (Is there a slide presentation for that?)

This paper examines the problem of surreptitious Internet interception from the eavesdropper’s point of view. We introduce the notion of ‘fidelity” in digital eavesdropping. In particular, we formalize several kinds of “network noise” that might degrade fidelity, most notably “confusion,” and show that reliable network interception may not be as simple as previously thought or even always possible. Finally, we suggest requirements for “high fidelity” network interception, and show how systems that do not meet these requirements can be vulnerable to countermeasures, which in some cases can be performed entirely by a third party without the cooperation or even knowledge of the communicating parties.

For practical results in real-world systems, see the authors' IEEE article, Signaling Vulnerabilities in Wiretapping Systems, in which this paper is reference #11.

In a separate work [11], we formalized the concepts of evasion and confusion as eavesdropping countermeasures and identified the “eavesdropper’s dilemma” as a fundamental trade-off in certain interception architectures.

The Eavesdropper's Dilemma - Matt Blaze et al... [PDF]