My gut reaction is "Verbatim copying offers no value added. Forbid that."
My post hoc justified gut reaction is that while there are sites on the Web which act as aggregators of information already present somewhere else, and do a great service to users, SA (and the other SE sites) isn't this type of site. Its strength lies in offering original information. Copying from other sites which (supposedly) also offer original content shouldn't be allowed.
More thought however reveals that tag wikis themselves are a very specific kind of content. They are definitions. The quality criteria for definitions are: a definition must be very precise, and it must be formulated as condensed as possible. Therefore, retelling definitions in one's own words is not a good idea. Science is a great example: when your paper bases on someone else's work, you retell (a summary of) his arguments, but cite his definitions word for word.
Consider a hypothetical tag chard. The Wikipedia article on chard starts with the description
Chard, also known by the common names
Swiss chard, silverbeet, perpetual
spinach, spinach beet, crab beet,
seakale beet, and mangold, is a leafy
vegetable, and is one of the
cultivated descendants of the sea
beet, Beta vulgaris subsp. maritima.
If you want to make a better definition, you need to state the same information in a more precise way, or in less words. I don't know about you, but I cannot think of a way to achieve either. I could rearrange the words and substitute a synonym or two to arrive at a reformulated definition which is as precise and as condensed as the old one. But such a definition is not better than the Wikipedia one, and doing so costs time without adding value. Worse: if I am an editor and find a very good definition on Wikipedia, but a "plagiarism" rule forces me to reword it, I may well end up with an inferior definition.
Of course, the purpose of the wiki is not to tell the world what chard is, but what topics are/should be covered by questions marked with the chard tag - which may be a deliberately chosen subset of the usual meaning of the word chard, or include an additional, site-specific meaning, or, if we decided to play a prank on the uninitiated site visitors, something completely different from the usual meaning. A software engineer would say that we are creating a project specific ontology (which is even narrower than the beloved domain specific ontology), whereas Wikipedia contains definitions which are as inclusive as possible (are they creating an ontology of the world?). A fully imaginary example: Maybe Wikipedia defines "Milk" as three disambiguated meanings, "The fluid produced by glands of female mammals", "A liquid produced by leaching ground nuts in water, like coconut milk" and "A white liquid secreted by some plants", but we want to only include the first meaning in our tag. Copying the three Wikipedia definitions would allow a question like "I am making dandelion salad, how to remove the bitter dandelion milk?" to be tagged with milk, and if we don't want that, we should make sure that our wiki describes exactly what we want to see in the questions and nothing else. However, I expect the majority of our tags to have the same meaning as the word in general, so good quality Wikipedia definitions should fit in most cases. Discrepancies could be removed on a case-by-case basis, possibly after a discussion on meta whether we actually want to allow questions about dandelion secretion to be tagged milk.
So from a purely teleological point of view, I am afraid that we should allow verbatim copied tag wikis, however ugly that sounds to my "always original contributions" attitude. Combining it with the editing motivation based arguments Aaronut lists in the question, it seems that it would allow us to create a tag wiki collection of higher quality (both in terms of the quality of the individual tag wikis and the completeness of the collection).
My argumentation until now assumes that we are dealing with a high quality copied definition. It is obviously a bad choice to copy a low quality definition. But we are not forcing anybody to do this. While banning copied definitions 1.Prevents the community from using some high quality definitions, and 2.Creates lots of bureaucratic work for a regulation body (the moderators?) to make sure that there are no copied tag wikis, allowing them doesn't mean that we will actually absorb the bad quality definitions found in some Wikipedia articles.
Look at economics theory: regulatory policing is much more complicated and fails much more often than incentive based policing (the current state of copyright law is a prime example of heavily used regulatory policing). So if we create a free "market", where copied definitions are not banned, and expect to see an unwanted side effect, such as the emergence of low quality tag wikis copied from somewhere, regulation isn't our only option. We can manage it through installing the correct incentives. We already have a very valuable asset: Our editors are intrinsically motivated, unlike paid employees. Even if the easiest way to create a presentable-looking result is to copy it from somewhere, they are interested in creating high-quality work instead of going the easiest path. So the only unwanted incentive in our system (that I can think of) is the one Aaronut mentioned in the question body: When there is no tag wiki, a user is likely to consider creating one. When there is a mediocre tag wiki present, a user (who is WLOG capable to create a better version) is less likely to consider changing it, either because he doesn't ask himself about its quality and thus doesn't recognize it as being mediocre, or because he is insecure about changing somebody else's work.
While I would say that the best solution would include a free "market" plus an incentive to improve not-so-good tag wikis (be they copied or not), I don't yet have a very good idea for such an incentive. A very simple idea (lifted from Wikipedia - I am aware of the irony) would be to label content which needs improvement. But first, an elegant implementation would probably require some change to the SE engine - probably minor, but I am not sure if it is easily done, or if it is desirable at all. Second, if somebody notices a mediocre definition, why doesn't he supply a better one himself? While there are legitimate reasons not to do so - and having just enough expertise to recognize a bad tag wiki, but not enough to write a better one is a legitimate reason in my eyes - this isn't a very good solution.
A better idea would be to somehow make good tag wikis recognizable as good ones. It isn't even necessary to require bad ones to be marked as bad - an editor should know that it is desirable to improve any wiki which isn't yet accepted as very good. This being a SE site, extending the vote system to tag wikis comes to mind first, but it isn't without its own problems. If the upvote applies to a specific version of a tag wiki which will be edited often to create new versions, what happens to upvotes which were cast for an old version? Also, do we want to make tag wikis so heavyweight that they get their own votes?
I am sorry I cannot provide a complete solution. If I have better ideas, I will edit the post. Until then, I hope my contribution will help others to come up with a better idea for the incentive. Or we can just say that we will tolerate the potential loss of higher quality definitions whose creation is inhibited by the existence of a low quality copied ones, and take no further action.
