CGI::Parse::PSGI::parse_cgi_output optimization idea

[bambr]

Hi Tokuhiro,
Thank you for your work on PSGI.

I had a problem with memory consumption on CGI::Parse::PSGI::parse_cgi_output and looked inside. I suggest to split support for filehandles and scalar refs into 2 separate subs and to stop using HTTP::Response. The code provided is about 13 times faster on small response (8K) and 15% times faster on large response (about 10Mb). Also the memory consumption is reduced on 15% on large response (and in becomes really great when we inline this code to avoid copying $stdout_buffer).
What do you think about this code/idea?
parse_cgi_output.txt