Chunking currently has a few performance issues compared to jboss web:

• The chunk end byte is written out as a separate write() operation, as the buffer sits above the chunking operation
• flush() will always cause a write(), so flush( ) followed by close() will cause two write ops, even if the issue above is resolved.

JBoss web worked around this via a second output buffer, below the chunking implementation (the socket buffer). I don't really want to have to double buffer unless it is absolutely necessary however.