In any case, I reached the limit of my debugging for this evening. I suspect either the content-length thing, the different string flushing behavior, or both. We'll have to dig into it more later.