The latest release of MEGAN has a “Min Percent Read to Cover” item in the Options->LCA Parameters dialog, Advanced tab. You could perform binning with a threshold set to 90%, say, then only reads with at leaset 90% coverage will be assigned to taxa, the others will go to the Unassigned node.
Then select all nodes except the Unassigned node and “Extract Reads” for those nodes.
Then you will have the reads that have 90% or more coverage.
Note that the “Min Percent Read to Cover” item performs slightly differently, depending whether you are in long read mode or not. If not in long read, i.e., if in short read mode (the default), read coverage is based on the longest alignment found for a read. So, in theory, if a read has two short alignments, side by side, then the coverage will be based on one of them, not on their union. This is for perform reasons.
In long read mode, coverage reflects the number of bases covered by any of the read’s alignments, which takes longer to calculate, but is necessary, of course, for long reads.