I went to the version of this talk he gave at Java Symposium; his argument was that it can work for Clojure, where immutability and functional purity resolve some of the problems. Perhaps these discussions moved his position slightly.
Or maybe he just learned Clojure people are pedantic and it's more time efficient to agree rather than argue with them.
Apparently Project Fénix https://fenix-ashes.ist.utl.pt/ at the Instituto Superior Técnico in Lisbon somewhat changed his mind about STM last May: http://www.azulsystems.com/blog/cliff/2010-06-04-or-how-i-go...
SPJ's take on Microsoft STM attempt: http://www.mail-archive.com/[email protected]/msg7973...
Or look at how STM is handled in Haskell, where it's arguably more pure.
http://lambda-the-ultimate.org/node/4070
(the optimistic vs. rollbacks distinction
http://liyang.hu/pub-model.xhtml
http://www.reddit.com/r/programming/comments/daqho/is_transa...
FWIW, why .NET can't have it:
(orig. post deleted):
http://www.reddit.com/r/programming/comments/6l7cb/clojure_s...
http://msdn.microsoft.com/en-us/magazine/cc163552.aspx