Actually, having looked at the processor implementations, there are just a few that need to do clean up. For example, in case of managed beans it's 1 out of 12. So, apparently, it's not that common.
Another thought on this subject is that if we are to add some clean-up method then perhaps it makes sense to invoke it when the unit is undeployed. I.e. go through the same deployment chain and invoke clean-up on the processors.