We want nofollow to be a succesful and useful initiative, which in at least some small place makes the Internet a better place to live. Nofollow will work so long as most search engines find the information contained in the nofollow tag to be generally useful on average.
If nofollows are sprinkled all over the web indiscriminately, then search engines which choose to ignore them will be at a competitive disadvantage to that which choose to respect them.
Therefore, what we can do to help ensure the success of the nofollow iniative is try to be sure that nofollow tags encode useful information for search engines. That is to say, links which are deemed *good* by *humans* do not tend to contain the nofollow tag, and links which are *suspicious* tend to contain the nofollow tag.
Therefore, it seems best to me if nofollow is on by default in the software distribution (since most small wikis are victimized by spam) and that nofollow is turned off in all the busy wikipedia sites.
Wikipedia provides an enormously high value set of hints to search engines to help them find sites that don't suck. We should not lower the value of the nofollow tag by labelling all of our high-value links as being suspect when they really aren't. Doing so reduces the value of the nofollow link for search engines.
--Jimbo