Unbelievably, I seem to have [hacked the version of long DEFINE-METHOD-COMBINATION from Sacla/XCL o pass the long form D-M-C ANSI tests][1].
I'm slightly incredulous that my patch works, because I don't totally understand the code I've integrated here. In particular, neither method combinations which use CALL-METHOD with the optional NEXT-METHOD-LIST, nor use of the MAKE-METHOD macro will work, although their implementation would not be terribly hard if I have the basics correct.
An aside: The more I work with our CLOS, the more I want to re-implement it from scratch, or at least separate the base out from the AMOP as a starting point to clean it up. Maybe this wouldn't help as much as I would hope, but reading clos.lisp is an exercise in frustration that doesn't end in admiration for clarity.
So, I hereby solicit additional source of tests or failing test cases. I can't promise rapid response to fix things, but I will get to them eventually.
If this work doesn't turn out to be too broken, I might commit to trunk in the next few days if the other developers acquiesce, with the rational that an incomplete long D-M-C implementation is better than none.
[1]: http://trac.common-lisp.net/armedbear/attachment/ticket/66/run-d-m-c-2010081...