I also don't see it as "p5p vs CPAN", because I would say that all participants on perl5-porters deeply care about keeping (modules on) CPAN working well and work with module authors to keep the breakage introduced through changes low. Bugward compatibility takes a backseat to correctness for example.
I agree; they're pretty rigorous at investigating how changes affect CPAN modules. That doesn't mean they won't make the changes anyway, if they're overall beneficial, but they seem to avoid module breakage pretty well.
An optimization for how constants were put into the stash in one of the 5.27.x broke the Type::Tiny test suite (and also Role::Tiny and Variable::Magic, I believe). They backed it out and worked with module authors, providing patches and additional test cases, before rolling it back into the 5.27.x code ahead of the stable 5.28.0 release.