The attack is not new. But it's still everywhere. That's why we're giving a talk at Blackhat. We are hoping this will finally be the year people take timing attacks seriously and fix them.
The point of our talk is to give concrete numbers to let people make their own decisions about exploitability. One result is a matrix of language (C, Java, Ruby, Python, PHP) versus attacker vantage point (Internet, LAN, VM-to-VM on same host). We'll show the minimum timing delta we were able to distinguish at each point, given a certain number of samples.
The previous best results in this area were 20 microseconds Internet, 100 nanoseconds LAN with about 1000 measurements (Crosby et al). We have improved a bit on this (too soon to give exact numbers, wait for the talk).
http://www.cs.rice.edu/~dwallach/pub/crosby-timing2009.pdf
Nearly every OAuth and OpenID library we found was vulnerable. No one has fixed these kinds of things. That's sad because they are exploitable in some configurations and as a software developer, you never fully know your customer's threat model. They could be running on Slicehost and have attackers literally on the same machine or a single server on a WAN link locked in a vault. (Usually the former more than latter.)
We hope our talk helps developers take this kind of attack more seriously but also dispel some of the FUD that these are easy. I think a good conclusion to make is "timing attacks are easier than I expected (but not easy) so it's worth fixing them."