Actually I disagree with this. A normal dyno measures torque at the roller, NOT at the wheel. The roller can put a lot more friction on the tyre than a normal road, and the number of rollers (1 or 2), where the tyre is positioned (on top or to the front or rear) and most importantly how and how hard it's strapped down can make a big difference. That's why the dyno does a coast down, where the speed is allowed to coast down while the dyno measures how fast it slows. That lets it calculate the drag caused by the dyno and the transmission too, but it's the dyno drag that's important.
That makes RWHP the dodgy figure, not the flywheel one. It's not a guess, it's a measured figure. Unless you're on a crap dyno that doesn't do a coast down. Incidently, that is the drawback of hub dynos, they can't do a coast down so for them the flywheel figure is a guess, and the lack of friction caused by strapping down a tyre is why they appear to read higher. They are more repeatable though, as the coast down can be a bit innacurate as the tyre may shift and the friction change when the car is under power.