I use SOAPdenovo2 to assemble a large genome (4.8G) using ~20X paired-end reads. The total length of contig sizes is 6.3G while total length of scaffolds is 2.7G. Note that this is a haploid genome, so there is no issue of heterozygosity for scaffolding.
I am wondering what happened during the scaffolding procedure? How comes that sum of scaffolds is so much smaller than sum of contigs?