While I've seen DNA seq being used to detect gene fusion events, more people seem to use either RNA-seq or cDNA-seq to detect these events. Is there some technical or other reason why RNA is thought to be better for detecting gene fusion events?
Breakpoints for most fusions are intronic ones, and introns tend to be very long. If you use gDNA you'll need to sequence them entirely. In mRNA of fusions those introns are spliced, so you'll just need to catch the abnormal exon-exon junction. You also are going to catch fusion genes that are more likely to be expressed in the cell. Software for fusion detection called deFuse (http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1001138) is even using expression values during fusion candidate filtering. This could also have drawbacks - e.g. if gene fusion puts 3' parter which is tumor-suppressor under a silent promoter, you'll not going to catch this likely oncogenic fusion.
In one word, depth. With DNA, you'd be randomly sequencing the entire genome and would end up spending a good bit of money to get sufficient depth for fusion calls. Since you're interested in fusion genes anyway, just looking at transcripts will reduce the amount of sequencing required.
While looking at genomic DNA would more easily allow you to look at structural changes, you can think of cases where no structural change is required for fusion genes to be produced or where it's unclear if a structural change will actually lead to a fusion.