Unfortunately, this excellent article is not yet publicly available on the web. I have excerpted those passages where leading scientists are quoted
"One can think of a quantum mechanical ensemble in many different ways. We showed that the pseudostates in present NMR experiments can be written in terms of purely classical probability distributions."
"When we think about quantum computation, we think it's weird and wonderful and it runs off the fantastic properties of quantum systems. These properties appear to be things like superposition and entanglement. So what can it mean if these machines have not produced any entanglement?"
"Since entanglement is generally considered to be the basic requirement for the success of quantum computing, this result raises very serious doubts about NMR pseudo-pure state computing."
"When people say this isn't really quantum, I couldn't agree more, in that we're simulating a pure state. But its evolution is governed by a quantum Hamiltonian, which for now lets us study the implementation of quantum algorithms. In the long run, this issue becomes irrelevant in the limit of strong polarization, a requirement that's been apparent from the outset."
Schack and Caves have explored the question of whether the dynamics of the pseudo-pure states can be modeled classically.
"We weren't able to do it. Maybe what that says is that something quantum mechanical is being done in NMR, but we haven't put our finger on it. What's definitely different from classical physics is that they're actually able to implement the unitary transformations that go into the quantum algorithm. It's just that they implement them on states that are always separable."
Raymond Laflamme (Los Alamos National Laboratory)
"If we do unitary transformations on the initial state, we can reach the answer in a much more efficient way than if we do only classical transformations."
"(For Peter Shor's factorization algorithm), the single quantum computing protocol known so far that allows for exponential speedup of computation relative to classical computation, entanglement is needed to achieve the exponential speedup. Merely producing a quantum evolution is not the point. What we are talking about here is finding ways to harness quantum evolution for performing computation better than with classical computers. This is what we claim is not possible with the pseudopure state technique."
"(In liquid-state NMR), the size of the ensemble has to grow exponentially in order for the system to work. If that doesn't happen, then very quickly the signal-to-noise ratio makes it impossible to see an output."
David Cory (MIT)
"For a quantum computer, the resources needed for a computation have to scale appropriately. Liquid-state NMR can certainly process information-the dynamics are great-but it does not scale as a quantum computer, so it's a mistake to call liquid-state NMR implementations of quantum information processors, quantum computers."
Andrew Steane (University of Oxford) warns, however, that NMR is restricted to looking at an ensemble average instead of measuring individual qubits. "This is the essential limitation of NMR," he explains. In addition to causing scaling problems, "it means that you can't do certain experiments where measurement is an essential part of the process, like when you're looking for nonlocal correlations."