I think it often comes down to what's the easiest asset workflow that gets us a good result. From that perspective, 1x raster + 2x raster is less work than 1x raster + SVG
Truetype fonts are an example of a vector format that can still look sharp at arbitrary resolutions because they can include hinting programs that align the control points to pixel boundaries. This is why fonts tend to look sharper on Windows and Linux than MacOS X. Linux and Windows default to strong hinting, and MacOS X uses only slight hinting, preferring accurate shapes to sharpness.
I know you already know these things, but for the sake of discussion (or other people).
Linux allows you configure hinting. It's true that by default (depending on how you define default) you get no hinting, but no reasonable distribution that I know of has shipped with that default for the past 7 years. Quite the opposite in my experience. Linux fonts (thankfully) look more like OS X fonts than Windows fonts.
Unfortunately, this configuration usually happens in some Gnome/Unity/XFCE/KDE configuration format, so if you want to use a better window manager with no DE, you sometimes have to recreate this from scratch, although, as someone who obsesses over font rendering and doesn't like the DEs, this is easier now than in 2000.
And then you have Google Chrome who just ships with its own font rendering. Gah.
For a quick glance, take a look at this nasty ultra-dense rendering in FF/Chrome and compare it to IE: http://projects.voanews.com/ebola-tracker/ (uses raphaeljs so renders in SVG or VML, whichever is supported.)
It doesn't -- it's the same as a bitmap in the default resolution rendered. And it gets even better when you zoom-in on those 'low resolutions'.
This is what we do for the cards at http://greenfelt.net. We lose a lot of detail in the face cards when we render them directly to the correct size as SVGs, especially at small sizes. There's a price to pay, of course, and it's a slightly more blurry look.
Another tangentially related but interesting fact is that every card set we have rendered as a pngcrushed png (except for the absolute largest size) is a smaller than the SVG source code.
Considering SVG is XML and PNG is binary, I am not totally surprised. Does that still hold after gzip compression?