I looked at randmly at some tasks from last year: http://www.google-melange.com/gci/homepage/google/gci2011
Some were given 4h, 24h, 48h or 96h. I guess that tasks of 24h would be a good estimate to plan for 24h (without assuming they will be programming non-stop, of course :).
Some tasks were quite open, such as "Find and report a bug in $product" or "Find a security vulnerability (a file that crashes the program)". I'm not sure if asking for something we don't know if it's doable is appropiate, though. OTOH they allow multiple instances of the same task. Others like update $PROGRAM to use this new API doesn't seem to have been successful. I don't think tasks of that level are appropiate (we could place a few, but without an expectation of being solved by the average teenager).
A good thing is that I don't think that reviewing a task will take longer than 5-10 minutes. So it shouldn't be a burden on the mentors. Seems a workflow like the one used on OTRS. The key is having a big pool of developers so a few ones don't need to review all of them.
An easy way to get articles would be in the documentation front, asking for a couple of wiki pages documenting something, a tutorial (with screenshots) on installing MediaWiki... Creating X new translations for MediaWiki or its extensions on translatewiki would also be an easy way of producing tasks.