Branch names with reserved URL characters in them, such as /, surface a bug in NameEncoder.decode(). Other characters can surface this bug too, such as ., .., and an empty space.
It appears to only affect jobs which were migrated, back in the timeframe of JENKINS-41124. This hasn't been recreated on a "real" environment with post-41124 projects.
With multibranch pipelines that have branch names including these reserved URL characters, such as the / in feature/branch-name, this method gets called as part of the creation of a file called name-utf8.txt, which defines the computed folders for the branch jobs containing these special characters. The bug causes this file to essentially double in size.
The issue lies here.