This piece does a good job of discussing the challenge of having collegiality as part of the criteria for tenure/promotion in universities. The stakes are, of course, very high, since tenuring a person who is pretty hostile/obnoxious means living with them for about twenty or forty years. On the other hand, as the piece notes, coming up with criteria for good or bad collegiality is pretty hard, even harder than measuring research contributions or teaching quality.
In my first tenure-track job, it was one of the tenure criteria, but given that the senior faculty were stacked with exceptionally un-collegial folks, the criteria could only be applied in ways that perpetuated hostilities. So, that is clearly a danger. The reality in tenure/promotion decisions is that even without collegiality clauses, the standards can be bent enough that a committee can recommend those that they want to recommend and reject those that they do not want to recommend. Not all tenure/promotion committees act that way, but it can happen [did it happen to me? um, sort of]. It is up to the rest of the university (and outside letters) to make sure that department committees do not bend the standards in this way. The collegiality standard will always be hardest to correct by those up the chain as the teaching and publication and service records can be reviewed. But assertions of asshole-ness? Unprovable and the folks up the chain may not correct for it as they defer to the folks closer to the candidate who know better.
On the other hand, I can see why folks would want not to tenure someone who is not collegial. Stuck for life. Lovely. On the third hand, folks may hide their true nature until after tenure, so their lack of collegiality only becomes obvious after tenure.
My general position here is that collegiality can be a tool to reject people who otherwise meet the criteria, and this tool is more likely to be abused than applied carefully and appropriately. So, I would prefer that this clause not exist. In my current and most recent jobs, it does not seem to be part of the standards, and that is a good thing.