We are currently grabbing the first article <link ... tag, which feedburner happens to list as the comments link. Listed last is the "alternate" url which is what you're expecting. The PEAR XML parser class we're using currently, XML_Feed_Parser appears to simply happily grab the first <link ... tag for each post. AIUI we'd have to rewrite the feed parsing with another parser class.
I suspect this issue would most commonly be associated with feedburner feeds.