The versioning in ezcontentobject_attribute comes with a very large data overhead in multi language functionality.
If a new language version is created the draft version contains only the new language. But if this version is published every other published language is copied, so that this version contains all published language versions.
With a large number of languages this will produce a lot of overhead, because to handle this amount of language you have to increase the number of allowed versions. This cause a lot of similiar entries from untouched languages.
E.g. If english is the initial language version which will then translated to let's say 10 other languages will produce 10 equal english entries in ezcontentobject_attribute. If we say we have an article with 10 attributes and 1000 objects this will sum up to 100000 unneeded database entries only in english in ezcontentobject_attribute.

The behavior should be changed to avoid this duplicity.