Please do not do this. This totally changes the way CR is used and I really don't see a need for it. If the assessment package has specific needs to relate different revisions please create an assessment specific table to handle it. This level of change must be handled as a TIP anyway, but I think there may be a misunderstanding in the assessment package use of the content repository if you think this is necessary.

In the assessment package we have questions that have some general data. But then we need to attach some question-type specific data and some display-type specific data which we now do with cr_rels in the cr since every information in assessment goes to the cr anyway.

When some data of a question is edited, a new revision is generated, but old mappings to that old revision should not see any change! Since the data in the mapped type-specific items can change, too, we really need a mapping from revision to revision.

Quite frankly, I don't see a problem in relaxing this thing a little bit...

Why not just add the revision foreign keys directly to your revision type rather than deal with a mapping table at all for this case?

It sounds like you're tying revisions tightly together, which, as Dave points out, is CR abuse, but if it's justified ... why not do it directly?  Or is this a many-to-one or many-to-many mapping (if true then a custom mapping table seems like the right thing to do).