You can clone with
HTTPS or Subversion.
Once the required file is not found by gem_original_require, RubyGems are going through all gems and tries to find which gem can contain the requested file. Unfortunately, it also goes through gems, which were already activated. However, for such gems, it makes no sense to look for some files to load, because their load paths are already available and the Ruby's require would be successful in that case. My simple patch  is trying to short-circuit the unnecessary search for files, and searches just the yet inactive gems. This patch should improve the RubyGems require performance, although I did not tested it. In theory, the more gems are activated, the faster should the require be.
Alternative implementation could be extension of the original find_by_path to accept optional parameter with :active, :inactive and :all flags and the search would be modified according to this flag.
I like it, but it needs a test.
Hi, the test can be found here: https://gist.github.com/1486087
However, I am not sure if the test itself shouldn't be protected by something like "save_loaded_features" which is used in require tests.
Improve path searching during auto activation. Fixes #227