one additional data point: we commented out the call to disk_log:blog() in our common print routines, effectively turning off all logging. the problem then goes away. so something down in there relies on some kind of configuration that we've apparently got wrong. i hope that helps narrow it down a bit.