To those thinking the EPA should just drive the cars, even that won't actually get very accurate results. Real-world fuel efficiency depends on so many factors that it would be impossible to reliably and accurately measure them all. For example, so-called hyper-miler enthusiast employ driving techniques to maximize their fuel efficiency. Conversely an aggressive driver could easily drop fuel economy in half. Then we have differences in temperatures, altitude, and terrain across the country.
So with that in mind, I think the current, 40-year old testing regime is probably still our best bet. It may not tell you how much fuel economy *you* will get, but since it's done under very controlled and consistent circumstances, it can give an indication to you how it will do relative to other cars. Honestly that's the best we should expect.
I fear we're going to meet the same problem with "real-world" emissions testing. I don't know of any car out there that can meet standards all the time. Take the cleanest car and get it to accelerate up a grade and it will dump pollutants. Or punch it off the light and you'll dump a lot more NOx and particulates than if you accelerate at a more reasonable rate.
In short, "real-world testing" is fairly meaningless. The only way to actually accomplish this is to have sensors and recorders on every car all the time and measure it and average it over time (and after the fact).