What are mate pairs? How does that help with sequencing?
Mate-pairs are essentially reads taken from opposite ends of a single fragment of DNA. This is useful for de novo sequencing, where they act almost like a single, very long read. The problem with random sequencing (both shotgun sequencing and current "next-generation" sequencing technologies) is that it's not obvious how a given read relates to the other reads you have.
Using the jigsaw puzzle analogy, mate-pair sequencing is roughly equivalent to being told that a given piece is 8-10 pieces away from another piece in a given direction. You still won't necessarily know exactly where to put either of the pieces, but it narrows the search space hugely, especially when you have one in place.
[1] Mate Pair libraries: http://www.illumina.com/technology/mate_pair_sequencing_assa...
[2] Paired-end sequencing: http://www.illumina.com/technology/paired_end_sequencing_ass...
It's often possible to tell how many times a gene is repeated because, statistically, you should get fairly uniform coverage of the genome (although there are lots of artifacts in the sequencing process that makes this less true). If you get significantly more reads from a particular locus, it's likely that it's been duplicated.
Glancing at the figures in the article, more than 65% of this "structural variation" seems to be insertions or deletions of less than 10 bp, with only 2% coming from elements greater than 1,000bp.