I am no expert, but it sounds to me like this headline is an oversimplification.

The abstract of the paper suggests that the authors intend to support the claim that "not all quantum systems can be simulated efficiently using classical computational resources". Essentially that it's impractical to simulate complex nondeterministic quantum behavior using deterministic algorithms.

But what about quantum computers? It would seem to me like their inherent nondeterministic behavior would be an ideal fit for simulating other nondeterministic quantum systems.

In any case, it doesn't sound like they found any real theoretical limitation, just a practical one.