I wrote the below in response to someone who e-mailed me about trying out our partitioning approach for metagenome assembly.
yes, the original partitioning approach worked only on low coverage data sets. The main reason is that highly connected regions (repeats, from biology; and some kinds of sequencing errors) are incredibly hard to traverse.
Our 2014 paper used a modified protocol, which included both digital normalization and a "knot removal" approach. This should work, but...
...since then, we've continued to refine the approach, but it's not something we're actively working on. A big part of the reason is that there are some nifty new metagenome assemblers that resolve most of the problems we set out to solve -- MEGAHIT is the one that we recommend.
The leading edge of technology in metagenomics keeps on moving. Exciting times ;).