M. Eric Irrgang added the comment:
Actually, it looks like performance concerns were raised as issues [2303](https://bugs.python.org/issue2303) and [2534](https://bugs.python.org/issue2534). For #2534, the issue was different.
For issue #2303, behavior-changing optimization was proposed. The first comment describes the cost of requiring a check to whether `__instancecheck__` is present (though I don't see what the comment refers to in 1b7f891f416830d0c46ca1c9e1bfe62f05cda655, which is the commit containing the cited r58099, as best I can tell). However, the discussion may have ended prematurely when the optimization https://bugs.python.org/file9742/isinst.diff slipped in through a different change.
See https://bugs.python.org/issue2303 for discussion.
For what it's worth, Guido (author of PEP 3119) did not seem to object to https://codereview.appspot.com/504/diff/1/Objects/abstract.c and no one seems to have discussed the apparent inconsistency with PEP 3119.
----------
_______________________________________
Python tracker
https://bugs.python.org/issue35083
_______________________________________