I got the PM, and have been thinking about this. There are two parts to this question: First, can we predict WHERE? and secon, can we predict WHEN?
Where is easier: quakes tend to focus on tectonic boundaries. However, there are nuances. Tectonic boundaries are enormously complex, and that complexity means that they're usually fuzzy--it's incredibly difficult to identify where the boundary starts and stops. It's not like you see in the movies, where there's a crack in the ground and that's the fault. That said, there are some generalizations that can be made. For example, deep quakes (like the 9.0 that rocked Asia a few years back) tend to occur in the deep slabs of subducted oceanic crust under convergent boundaries. They tend to be a good distance from the surface of the boundary, but they're still part of it. Shallow quakes tend to cluster around the surface expression of the fault, as they're caused by the deformation of the surface rock by the fault. They tend to be more minor (though they can still be pretty powerful), since they don't involve slabs of rock that weigh as much as Japan (or more!).
You can also get earthquakes in ancient faults, which is what happens along the USA's east coast. Some of the faults were due to the formation of Pangea, and then the later rifting that formed the Atlantic; some are, as I understand it, much older. Those areas can be re-activated by various processes, and are difficult to understand because they're between the deep epicenters and the shallow ones, and far from any current tectonic boundaries.
Pull-apart basins such as the Basin and Range can generate earthquakes because, well, it's the start of a tectonic boundary. As the continents pull apart, normal faults occur as huge blocks of rock move downward relative to everything else.
Hot-spot volcanism can generate seismic activity as well, for exactly the opposite reason as the pull-apart basins: huge blocks fo rock are moving UP relative to everything else, sometimes quite explosively (Yellowstone, for example).
When is much, much trickier. I suppose in theory we could calculate it if we knew all the factors--Poison's Ratio of the rocks involved, strain rate, pore fluid pressure, Sigma 1, 2, and 3, all that fun stuff. Unfortunately, the only way to learn all of that is to excavate the feature, which destroys the system (or at least substantially alters it). And if the stresses are great enough less-intrusive methods like drilling won't work, because the pressure of the rock will deform the drill bits. The hole in Russia being drilled to the MoHo Discontinuity ran into that problem--couldn't keep the bits the right shape. And those are expensive buggers, too.
There are ways to predict quakes. For example, a large quake will almost always trigger smaller aftershocks and even large-scale quakes. That said, these can happen weeks or even months after the main quake. Earthquakes shift the system, releasing some stresses and generating others, and it takes a long time for the system to re-equilibrate sometimes. And if you've got numerous small quakes happening that may be a predictor for a larger quake--though this is by no means certain.
The tl;dr version is that right now, at the level of technology we have, it's simply not possible to predict earthquakes with anything resembling accuracy. We've got a good sense of the where (except when things hit us out of the blue), but as far as when goes it may as well be entirely random as far as prediction is concerned.
There's another geologist on this site you may also want to talk to: Correa Neto. He's better versed in hard-rock geology than I am, and may be able to provide more specific data.
