When removing a job A, all jobs B1, B2, ... are changed to remove A from parameterized trigger build steps and the publisher.
This makes perfect sense for the publisher, and "asynchronous" build steps, as the success of A does not depend on them.
But if the build steps in the Bs are configured to wait for A, and even to fail if A fails, then the build step should not be silently dropped from B1, B2, ..., as it can be assumed that for the Bs to succeed, A is required.
Users are then left wondering why the Bs don't behave correctly, and what happened to the build step. Leaving an invalid, always failing build step behind in that case would make more sense, as it'd raise awareness of the inconsistent job configuration.
FWIW this would also facilitate archiving and unarchiving of jobs like A without breaking Bs that are only around to provide e.g. their artifacts.