My concern with that approach would be that bivariate shapes used in this manner are notoriously bad at relaying the appropriate information.
Varied widths of bars cause a great deal of visual clutter to the display and distracts from all of the data - it poorly encodes the confidence value of the data that it is meant to display, and it also obscures the original purpose of the chart at the same time.
More often than not, encoding data in this way will confuse and obscure rather than enhance.
Two charts are quite often better than one at telling the story of the data.
This is also the type of data that error bars were intended to address, though that feature is not available in Highcharts either.

(don't mean to derail or hijack this topic...it's just a subject very worthy of discussion IMO)

@Jamie: That's definitely better than the dual-axes bar charts that I see a lot; if our project can't get the kind of chart I suggest, I'll recommend going with the small-multiple bar chart. However, I do think that bars with variable widths would be an intuitive way to show this kind of information--bivariate shapes for bivariate data. The widths would be relative, of course, so the precision of the data would not be immediately visible...but the _sense_ of the information would come across instantly. IMHO.

Sounds like the perfect job for small multiples: http://jsfiddle.net/eddcu/2/
Data encoded this way is much easier to interpret and understand than it is by changing the width of bars
FWIW

In a bar or column graph that ranks a set of elements (say, as a series of percentages), the thickness of each bar or column would indicate the relative robustness of the data on which those percentages are based. For instance, if one model of car had a better safety record than another, it would have a higher column, but if its safety record were based on half as many tests as the other car, its column would be half as thick.