Hello Professor Kosara,
I have extensively used both services and I agree with most of your remarks.
Here at the OECD like many "progressive" official statistics providers, we rethinking the way we disseminate data. So my focus is rather on "publishing data" than on "visualizing data".
In that aspect I am a bit disappointed by both offerings. In house, we store a great amount of metadata on any of our products or statistical objects and we're still far from perfection. But here, we can only fill a very limited number of fields, sometimes with strong constraints (i.e. no formatting in many eyes descriptive texts! not even line/paragraph breaks!). As a result I find that as a user (rather than as a data producer) navigating among the datasets requires considerable effort.
I've published a few hundreds datasets on both sites, and unsurprisingly a couple receive all the traffic while the others get none. that's simply because the successful datasets were on the top of a list at some point while the unseen datasets would have required the user to page down through results. once you have a rough idea of what you are looking for, it is not possible to drill down to it step by step, you must search for it exactly or never find it. (or maybe it is possible but then it is implemented in such a way that users don't do it). I see this as a big obstacle.
about the variety of visualization types available, what I've seen from IBM suggests that many eyes could implement many more, while swivel stick to a few basic ones. is that a bad thing, I think not, it is already so easy for a novice user to use the wrong chart type and produce meaningless results that still look relevant at first glance...
regarding technical choices, I also very much question the use of java. No, anonymous, java is not lightning-fast on all current computers, it is annoyingly slow on quite a few. I do second Matthew's comment about flex, which seems to be the obvious evolution!
Thanks for the article
-jerome