Is this written somewhere? I would agree that only a release manager releases new versions, but i would not agree that it is a good idea to cause inconsistent behavior this way. The question comes down, what "release" means. There should be a way to add updates to the source code repository, which can be tested, without "releasing" this version. If this requires updating the version number, there should be a way to update version numbers without "releasing" it. People who have initially installed 5.5.0a1 before your change see a different behavior than people initially installing the same version after your change. The problem is that even after an update from cvs (and a restart of OpenACS), the behavior is inconsistent. Essentially, a developer with frequent updates is able to test your change only after the version is bumped (which did not happen so far).
Anyhow, this does not belong to the topic of the thread - when it does not happen so often, i can live with it.