In the demonstration the MS05-025 patch is dissasembled and the vulnerability that was patched was discovered in less than 20 minutes. Reverse engineering patches isn't a new idea, I first mentioned it here last year and have even done it myself.
It does raise some interesting issues, particularly for the 'vulnerability research does more harm than good' camp, if you can reverse engineer the patch, then not releasing (most) vulnerability details will harm the administrator, as less information will be available in patch management decision making.
Are these patches really reducing the pool of available vulnerabilities or are they generating more holes than they are fixing by encouraging vulnerability research. I would argue it is mostly the former, but not always, and I don't think this topic is as dead as some people claim it is.
Then as the OSVDB blog points out, this may provide some usefull stats on which blocks of code are producing the most vulnerabilities by keeping track of which functions are being patched. This may provide some information to administrators as to which services are dodgy and information to vendors as to which blocks of code they should rewrite.