The problem doesn't necessarily have to be anything to do with bad packets but rather with available bandwidth. Are there any bottlenecks occuring at the time of the disconnects?