Agreed. It's possible to make it extremely difficult to walk the blockchain in the linear, simplistic manner I describe.
My point was that it's possible to make it *impossible* to prove the flow of funds, not that it's possible to make it difficult.
I am aware of the existing blockchain analysis that's been done by various researchers and how far it can go. But that method will never work against mixed inputs with mixed outputs, for the reason I described in the previous post. It could make a probabilistic connection between two addresses, but those probabilities will dwindle to zero after a few transactions, if they're all implemented the same way with N users.
You're more describing how it works now; people mostly don't do anything to hide their tracks, and some people use mixers,which *largely* work in hiding, but not perfectly.
CoinJoin I believe is just a slightly more sophisticated way to use the existing Bitcoin transaction design (honestly I'm a bit surprised that it's only being discussed now and not 2 or more years ago - I think the problem is not the idea, but that there's SO much functionality to develop within Bitcoin and so few skilled people to do it).
All fair points. I was, indeed, talking in the video about "as-is" rather than "could-be". And I was also implicitly assuming that there will always be enough people who use the system naively or employ lax security practices that there will be a "way in" for sufficiently motivated law-enforcement agents.
I also think you make an excellent point about the power of compounding/exponentiation with chained N:N mixing... you rapidly reach a point where the set of potential suspects is the set of all Bitcoin users. If such an approach were widely deployed then my statement that Bitcoin is more traceable than cash may well prove unfounded. But I sense we're a long way from there.