Working memory, sometimes called short-term memory, is currently thought to be closely related to IQ test results (for example) and perhaps to the cognitive performance that IQ tests try to measure.
It's also generally assumed that IQ cannot be improved, that individual capacity is determined almost entirely by genetics, intrauterine life, and perhaps the first few months of postnatal life. . On the other hand, there's some evidence that working memory can be improved by training, though we don't know if the training effects persist.
That's roughly where the published science seems to be, but we're always free to draw some speculative (ok, breathtaking) inferences.
Wild speculation number one is that for most professionals under 50 it's not worth investing a lot of effort into training or improving working memory. My hypothesis is that in healthy brains things are pretty much aligned and that there aren't huge differences in subsystem performance. In other words, there aren't big performance bottlenecks. Improving working memory by, say, 10%, might only boost cognitive performance by 1%. Physical exercise is likely a far better investment than short-term memory practice. (I'm amazed by how important physical exercise seems to be to memory preservation, I never expected that.)
Once you get past 50 or so the rules change. I think the majority of us have some subsystem failure by then, so gains might be larger. I won't explore this one, however, as my major interest now is younger persons.
Consider the child, or young adult, with mental retardation (old term), cognitive disability (newer term), pervasive developmental delay (somewhat meaningless term), autism-spectrum disorders, severe ADHD and the like . They often have patchy cognitive function. Some subsystems might work pretty well, while others, like working memory, might be pretty much shot.
This brings us to wild speculation #2. Imagine that IQ was the measured result of interactions between working memory, processing speed and, "rhubarb". (For this discussion "rhubarb" doesn't matter.) If "rhubarb" was working at 80% efficiency, and working memory at 10%, then working memory would be a performance bottleneck.
In this case a 10% improvement in working memory might become a 7-8% improvement in overall cognitive performance. That's a pretty significant improvement for someone who's on the borderline of social or quasi-independent function.
Now, that's an awful lot of speculation, but it's not hard to imagine how to test the hypotheses. We'd identify a set of children with low IQs, and work on short-term memory training...
Oh. Wait. It's been done. It worked. There are similar articles, including a review of the topic published in the PNAS that pointed out this is only the start of a long research agenda.
In the meantime, what about those low IQ children we know of today?
Well, when I wrote about this topic a few months ago Martin Walker of MindEvolve mentioned he's developed a software program to apply Jaeggi et al's techniques at home . I've also previously written about games for improving focal cognitive weaknesses, and there's a real cottage industry of brain training exercises, including for working memory . On my iPhone there are games like "Garf"  and "Matches".
I'm going to think about this a bit more. There might be a way to work this into our regular homework times for one child in particular ...
 In the interests of simplicity I'll assume IQ tests are a good approximation to IQ. It also obvious that IQ can be lowered by poor nutrition, head injuries, infection, seizures, trauma and so on.
 One of my pet peeves is the futility of classifying cognitive disorders given our limited knowledge of the underlying pathophysiology. Don't get me started on "reactive attachment disorder".
 As I wrote above though, I suspect those will have limited effects on the average "balanced" brain.
 My wife is very good at this. I am astoundingly bad at it. Hmmm.
 Windows only, $50, no trial version. The price is reasonable, but we don't have much Windows left in our home, and it's hard to spend the money without testing it first.