Just closed JENKINS-30815 as a duplicate of this one, after Carl George pointed it to me. But you can still have a look there for more evidence of how undesirable this compression can be in some cases.
Fixed in PR#4205 and released in Jenkins-2.196.