It isn't a matter of how it 'should' work. The current system is an approximation of approval voting (in fact, under true approval voting the number of votes would be unlimited, but this was changed not for any governance reason but due to an exploit concerning backup witnesses).
Approval voting has the property of electing the witnesses which are approved by the broadest and maximum amount of voting stake. Under most conditions this would be good, as it sets the maximum threshold for an attacker to come in with a narrow amount of stake and elect witnesses contrary to the broader consensus.
In this case, what people are trying to accomplish is not to represent the broadest and maximum amount of stake, but in fact handicap the largest stakeholder (who in fact has an absolute majority relative to actively voting stake). In a sense, this is exactly the opposite of what good, secure, governance would do.
That being said, it may be a reasonable tradeoff (lower security for less majority control) to limit the influence of the largest stakeholder in this specific situation.