Unfortunately, metagenomics is currently the lowest hanging fruit of genomics/biology right now. Sequencing has gotten exponentially cheaper to the point that these types of projects are feasible (the paper you cite is actually a proof of concept for their metagenomics pipeline), that everyone is just sequencing this shit out of everything. That said, there are quite a few pitfalls with this current approach:

1. Big data in biology is still lacking comparable analytical capability.

2. Most high throughput sequencing technologies (next gen sequencing) is using polymerase based technologies that have short reads. Polymerase (the basis of PCR) requires a primer to start the reaction - and the choice of primer can introduce non-trivial bias into the obtained results. Short reads are pretty good in dealing with most bacteria, but the reaction gets pretty hairy with hyperthermophiles (bacteria with regions high in GC content) and more complex eukaryotes (long stretches of repeated regions are difficult to resolve with the shotgun approach).

3. We will probably only see DNA reads that are slightly different from what we know to exist already. The generalization they make that 99% of species cannot be cultured with current methods is true - but an analogous scenario with DNA can be made. The sequencing reactions will probably be optimized for sequences with similar composition to what we already see. The task of extracting DNA from bulk material is not trivial - this can remove large swathes of species from view even if the sequencing is trivial. Complex 3D RNA and DNA structures such as hairpin loops and noncanonical pairing makes sequencing hard.

It's definitely still a difficult question to tackle, but more difficult than doing it are answering the questions "What can we learn from this?" and "What should we sequence?"