This is an important issue to consider, but talking about "pixels" alone doesn't get us far in this context.
For one thing, different fonts have different design characteristics. In particular, the x-height of one sensible body font at a nominal 16px could easily be 50% more than the x-height of another. Sadly, while CSS lets us specify a stack of fonts to try, it doesn't yet let us specify a different size to go with each one so we can achieve a similar optical result. (CSS3's font-size-adjust might be intended to help here, but seems overcomplicated and underpowered to me.)
For another thing, these days pixel densities can vary by at least a factor of 3 between devices. The latest iThings sometimes do better than the 300dpi that used to the be benchmark for a laser printer. For a classic CRT or cheap and cheerful TFT, something more like 96dpi is common. At least you can detect this to some extent in CSS, because the new devices with very high densities also tend to support media queries so you can present larger text (in pixel terms) and higher resolution graphics accordingly.
If we're going to improve the way we present text (and graphics) on the web for people whose vision isn't perfect, it's going to take a much more flexible styling system for the web and probably a lot more real world experience of what works and what doesn't as well. As much as I respect people taking a stand on behalf of those who can't see as well as some, I'm not sure advocating a fairly arbitrary 16px guideline is helping.