You guys were right.
My business process calls for concurrent HTTP posting to a partner. The broken pipe errors surfaced only after moving from NT development environment and into Solaris for testing. I have since staggered the HTTP posts and asded some REPEAT on FAILURE capabilities to minimize the instances of this exception.
So far, it seems to be holding. I will let you know if this work-around falls apart.