Regression tracking: unnecessary reference tasks?

From https://debusine.debian.net/debian/developers/work-request/572100/ (upload-to-bullseye-security: perl_5.32.1-4+deb11u5) it seems like all the QA tasks are run on the previous/reference version, independently of whether the new version succeeded.

This concerns autopkgtests and basic lintian/piuparts/... jobs, only reference blhc was apparently skipped with no subtasks.

Some reference tests are already completed while tests for the new version are still pending (here e.g. abi-monitor autopkgtest).

This sounds like a waste of resources, especially in base packages like perl.

I'd expect a reference test to be scheduled only if that test fails with the new version. If the new version works, I'm not particularly interested on whether this also worked previously. As a QA tester, I'm interested in regressions, not improvements :)