This could only work for annoyance and not for security purposes. The servers cannot practically reencode video on the fly for each client, causing the limited captcha randomness you suggested. Sending down a smaller overlay for the client to render would make it easy to actually find a frame where the important letters in the captcha are clearly distinguishable.

A plugin with database looks like it would be the way to go to combat this. From the samples on the nucaptcha website, the message is bright red moving text embedded in the mp4 encoded video. They may just have a hard-coded demonstration, but since it wasn't actually overlayed, it should be easy to look up the ID (or URL) of the video file, or even a hash of some small amount of the video sequence itself.

The samples I saw were each over 1.5 megabytes. I would be glad to to fight this waste of bandwidth.

Here's a sample video (warning: large) link

I am a MS hater as much as the next guy, but this is complete nonsense.

Yes, this is a PR stunt. Yes, it is intended to influence people by making them believe that MS has changed and is more open when it is not likely to happen. No, providing information about an older format that a user may still be using can not in any way be taken as a way to strip people of their rights to view the contents of their data.

That said, the article made me feel a little ill. I'll paraphrase it in much shorter terms:

• The data was already available by sending off a request by email, MS is just making it easier to get.
• There are many implementations of OOXML out there. [The provided list is questionable as far as completeness of implementation. I suspect most of the implementations are provided for document type detection or simple text conversion and not much else.]
• MS claims to be going above and beyond their duties to open a sourceforge document translator project.