Do you have a link to the paper where this was proved?
Suppose we take Tor and throw a bunch of fiber at it so that the bandwidth improves, then whenever two Tor nodes make a connection they send data at a fixed rate, sending real data when there is any and padding when there isn't. That clearly isn't "impossible" so what's the attack that works against it?
A naked insecure connection through HTTP has response times as low as 100ms coast to coast (US). This makes these connections look instant or almost instant. It's impossible to achieve neither anonymity nor encryption through a naked single connection - you need several, at the very least for AUTH - and this alone multiplies the latency vs a single connection. The problem is that the naked eye can notice the difference between 100ms and 200ms, so the experience cannot be the same. The speed of light is already significant in world-wide distances when you consider a number of hops and a minimum number of messages being necessary to ACK, AUTH and then provide some anonymity layer that hides both sides of the connection.
Anonymous, secure, encrypted random web browsing CANNOT be as fast as a naked open connection. The difference can be minimised over long transfers, but latency over many requests is always going to be several times higher.
You can make all connections theoretically so fast that even a 5-fold slow down is still "fast enough", but for many small transfers it simply cannot be done so fast that it's humanly not noticeable.
Alas, there's ALWAYS going to be a noticeable difference between open connections and hardened connections (to the standard we are talking about in this thread). A simple AUTH cannot be done so fast that a connection across the Atlantic is not humanly noticeable. And what we are talking about implies a whole lot more than that.
In addition to that, you can mask latency with caching and prefetch. Browsers could prefetch all the small content on linked pages so that by the time you click on a link the data is already on your computer. This could be helped by general purpose HTTP protocol improvements that allow a browser to know speculatively what such content will be on a page through a POST request etc. without having to actually POST anything. SPDY/HTTP 2.0 also allows asynchronous operation which is extremely good at masking round trip latency when requesting many small objects.
The post I was replying to wants casual browsing that's secure, anonymous, private and encrypted AND indistinguishable in speed from what he does without all these features. That's impossible, more requests are needed and every single connection adds noticeable lag past a certain minimum distance.
Some usage will have to be adapted and/or compromised will have to be made, or else just accept being monitored.
Overpromising is not a good policy.
If you want anonymity, privacy and security you will have to make certain compromises to your "current internet" that isn't anonymous nor private, but instant and convenient. Or you can just give up and accept having your every step in the net monitored.