This can be used to make a user's fingerprint stand out based on their browsing patterns. However, it is very fragile in practice. The tracker would need both a rare fingerprint, as well as a rare browsing pattern in order to identify a user.
This is pretty hard, considering the Tor Browser does a good job at having a common fingerprint at it's highest security setting (Javascript disabled, which is what this tracker is for).
I think he is saying that users can't be tracked between page-loads using this method, or your risk sending multiple users the same token. (which is true, at least with this implementation)
The time they spend on the website, latency, etc can all be used to add to a fingerprint, but there isn't something magic that makes this accurate, especially without JavaScript.
I may be missing something, but it seems to me that this technique(if not this particular implementation) could be used to easily track individual users.