The core problem boils down to the exact same reason why the Wii's interface is terrible: it forces you to make much larger, slower motions -- motions that can easily be misinterpreted and have very low accuracy -- instead of spending 0.05 seconds tapping a key or clicking a button.
Tablets are another fad like 3D movies. They're popular with a lot of people, but in the end, they just give everyone pounding headaches.
The fact that the interface is a fat finger instead of a mouse does not magically make the interface any easier to use. What mobile devices (not just tablets) have demonstrated is the power of simpler operating system user interfaces -- not that touch screens magically make everything a million times better. This will be the primary legacy of tablets, not the touch screen.
From my experience with iPhone/iPad, the people who "think" that the touch interface required precise input are the ones that end up with lots of spurious inputs, because they use very light and "careful" touches that are misinterpreted instead of letting the input prediction do its work.
I do agree that there is a lot to be done regarding keyboard input, but even the keyboards seems to work OK for most people.
Touch interfaces actually do work pretty well though. Direct manipulation is just more engaging.
This fantastic video comes to mind: http://10gui.com/video/
It's not yet clear to me if touchscreen will ever make sense for larger-screen devices, certainly you do not want to interact using touch with a vertical surface for any length of time.
However, using an iPad for "long-term, serious work" is just fine. So I don't think you should equate large screens with serious work.
I think the 10GUI idea linked above is great because it's intuitive enough and still robust. The intuitiveness and engagement of the iPad is great, but I find it restrictive (the input bandwidth is limited by the direct manipulation paradigm). I think it's fantastic for people who have trouble understanding or don't particularly feel the need to spend time learning the intricacies of indirect manipulation, but I don't see that as a problem for my generation (cusp of X/Y) or subsequent ones.