I think your analogy is correct, and vector cutting is essentially identical to a waterline approach with a fixed tool diameter (the laser spot size). So if you have a colored or layered vector regardless of the format it is derived from you can assign power (and speed) values to each color/layer so that the result will be varied in depth. You can define cut paths (follow this line) and engrave paths (cut this entire area at one power level) and off it goes.
Rasters are different since a raster image is just a matrix of data (i.e. pixels) and each pixel can only carry a numeric value. When engraving a raster it appears that the Glowforge software will simply apply a power level to each numeric value (and interpolate down to get a maximum of 256 levels). So, say you have a grayscale image, then the laser applies maximum power to pure black pixels, no power to pure white pixels, and power relative to the darkness of gray for everything in between. It is not clear to me from watching the available video if it does so in multiple passes (i.e. cuts the same area over and over to achieve the final depth in a waterline approach) or if it pulses the power level to approximate the color at each individual pixel. It will be interesting to find out as more information becomes available.
I can say that when I was at the open house I watched the scanned items being engraved, and the software was clearly taking the raster from the camera and thresholding it to get a solid area. My take is that the software may be doing a smooth and threshold kind of operation to reduce the levels of gray down to 255, and in the demos the threshold was set to just two levels (i.e. black and white). The software that is generating motion planning from rasters also seems to have a notion of “cut the outline when you are done engraving” by following the outermost detected edge.