[krizhanovsky]

> I'm unable to find measurements. I mean, the amount of code duplication is not an optimization parameter, the amount of Mb/s of HTTP parsed is.

The measurements are covered in slides 23-24 in http://www.tempesta-tech.com/research/http_str.pdf

> goto jumping between branches is a bad thing from the point of view of maintainability.

In our parser we have DSL for the state machine, so we encode which state the machine should go. The HTTP parser is one of the most updated piece of a web accelerator code and we don't struggle on the FSM - every developer in our team updates the code, even newcomers.