Even if you remember to implement a canonical tag pointing to the "real" page, you risk people linking to these dummy pages (producing a minor loss of link value according to Cutts) or weird mangled versions coming into existence (think escaped referrer logs that make their way public and crawlers find, etc), and then bots have to take the time to request incorrect URL's and find out they're junk. Better to return a 404 or 301 to the correct page (if you know what it should be).

See, this is actually the kind of sh*t that SEO is about: usability for bots.

If you're going to have slug-type Urls, they should at least be unique, and if they do change, the old version should at least show a canonical link and/or 301 to the new one.

Usually this means keeping a record of every variation the Url has ever been for a piece of unique content.