A streaming token parser might well get you good performance for that format... maybe something like an antlr grammar or even a simple scanner. Regex is not the only pattern :)

It would also be great to see such a parser contributed back to the community of possible, and I sure we would be happy to help maintain and improve it in the open source.