I know you say that in jest, but I've always been curious if anyone has ever compared the size of tokenizing an OS X Quartz 2D stream that renders a screen against a PNG of the same screen. I wonder what we lost on the road not taken when as an industry we solidly branched away from Display Postscript as a possible industry-wide rendering engine, and it is unlikely we will ever explore that branch in the foreseeable future.
As noted in a thread comment elsewhere, under the hood our display rendering engines are already entirely or significantly vector engines, but what we appear to lack is an industry-wide representation scheme.
That's an excellent point. Screenshots are currently the exception, but shouldn't be - most UI is vector but OS screenshot tools don't record them properly.