As to the second point: Who said anything about _me_ running a javascript eval on user supplied data.
Go look at their wiki example click edit and in the text box type: <script>alert('boo');</script> then click the hide button.
Notice two things. The first is that when you type the last > on the ending script tag the included script executes and you get the alert popup. When you reload the page you'll still see the alert('boo'); in the wiki display (in <code> tags) but the <script> tag is still live in the page as there's either no or very limited output sanitization. In a typical client/server application this is hopefully going to be sanitized or eliminated prior to ever being stored. The storage format is irrelevant if the data being stored is spit right back out into the page as in the wiki example.
But that is indeed a good argument in your favor, ktsmith.
As client-side JavaScript gets more mature, I do expect there to be more frameworks that will just "do the right thing" for developers who write code like this.
For you second example, why the heck would you run javascript eval on the document data? As explained by mckoss, the storage format is JSON, not raw javascript that is executed. This is an important distinction: jQuery and other client libraries contain special commands for handling JSON that exist specifically to resolve the kinds of problems you are describing.
EDIT: It looks like the other reply explains that there is no "trusted agent", so this is clearly a limitation of their system (though not a theoretical limitation, as far as I can tell- No reason they couldn't allow the app creator to log in with pseudo-admin privlidges to do reporting.)