First I would make sure that you have no other homologies in your primer or the kan cassette elsewhere in the genome. Secondly I would check that your method of screening for integration events is actually correct, perhaps your analysis is not quite right, or maybe you are getting integrants instead of replacements. 

For screening I used the forward and reverse primers that I used to construct the cassette from the three fragments (i.e. the forward of the upstream arm and the reverse of the downstream arm). If bcsQ has not been knocked out, using this primer set I expect a fragment of 1.5kb and 2.3kb if the cassette integrated correctly. I consistently get a 1.5kb fragment. I also use primers that are internal to bcsQ. If bcsQ has been deleted I expect no amplification, if bcsQ is present a 440bp fragment, which I obtain with all my transformants. I also used a primer set taht is internal to kanamycin, and all my transformants test positive. The last primer set I use is a reverse primer that lies close to the start of kanamycin and the forward primer that is upstream to the forward primer I used to generate the cassette; using this primer set I get no amplification, indicating that the cassette did not integrate in the desired location.

I checked the homologies of the cassette against the shotgun genome; using the lowest specificity, I got 4 regions that matched with 16bp, 16bp, 17bp and 20bp homology.