Dennis Sosnoski has posted XML Binary Infoset Encoding (XBIS) 0.9, a Java class library for converting XML documents into non-XML binary documents that he claims are faster to parse and generate be factors of four to eight times. However, I don't believe those numbers for a minute. Sosnoski's benchmarks are comparing apples to oranges. His SAX numbers use parsers that perform well-formedness checking, and the time required to check well-formedness are included in the input time. However, the XBIS parser relies on an underlying SAX parse to check well-formedness before the XBIS input data is generated; but the XBIS numbers do not include the time spent on this parsing including well-formedness checking. In fact, I would not be at all surprised that if the well-formedness checking cost were also included in the XBIS numbers, that it would prove slower than real XML parsing. There's also some question about how well optimized and representative the SAX output code is. XBIS is published under a BSD license if you want to check it out for yourself.