It seems that the new TrackBack auto-discovery feature downloads each URL you link to and parses it to figure out how to send a trackback ping. The problem is if you link to a large file like an mp3 it downloads that and then if your CGI provider puts limits on your CGI's memory usage you'll run out of memory and get a 500 internal server error. Best solution I've found so far is to turn off the Trackback auto-discovery feature.