Benchmark each require in your Ruby code.

gix_require_benchmark.rb

require 'rubygems'
require 'rbtree'

class Req
  attr_accessor :path, :time, :requires, :exception
  def initialize(path)
    @path = path
    @requires = []
    @time = 0
  end
end

module Kernel
  @@require_stack = [ Req.new('<top>') ]

  alias original_require require

  def require(path)
    req = Req.new(path)
    @@require_stack.last.requires.push(req)
    @@require_stack.push req

    result = nil
    t1 = Time.now

    begin
      result = original_require(path)
    rescue LoadError => e
      req.exception = e
    end

    t2 = Time.now
    req.time = (t2 - t1)

    @@require_stack.pop

    unless req.exception
      result
    else
      raise(req.exception)
    end
  end

  def print_require_tree(req, loaded, level = 0)
    indent = '  ' * level
    fmt = " #{loaded[req.path] ? '(%10.5f)' : ' %10.5f '}  #{indent}%s  %s"
    puts fmt % [ req.time, req.path, req.exception ? "(#{req.exception.message})" : '' ]
    loaded[req.path] = true
    req.requires.each do |r|
      print_require_tree(r, loaded, level + 1)
    end
  end

  at_exit do
    print_require_tree(@@require_stack.first, {})
  end
end

require_benchmark.rb

require 'rubygems'

module Kernel
  @@require_times = {}

  alias original_require require

  def require(path)
    t1 = Time.now
    result = original_require(path)
    t2 = Time.now

    @@require_times[path] = (t2 - t1)

    result
  end

  at_exit do
    @@require_times.sort_by { |path,t| t }.each do |path,t|
      puts "  #{t}\t#{path}"
    end
  end
end

One suggestion I'd have for require_benchmark.rb is that you only track the require times when result is true.

If you mistakenly require something twice, the second time the result will be false and you'll overwrite the original time you tracked. Since a second require should be lightening fast, due to ruby caching the fact that it was required, it could be changing the benchmark to make it look like the file loads way faster than it does in reality.

Fixed. The only failed require appears to be fastthread which incurs ~0.09 seconds.