There is no such thing as an accurate rolling road, especially if you have the figures in flywheel (which is just a guess).
If you guess at the losses of course your result will be a guess. If you measure them accurately using a coast down test and the dynamometer itself is calibrated you'll come close to the results you would see on an engine dyno.
Most rolling roads work on the "Guess" principle however, IME. Either that or the "tell the customer the figure he wants to talk about down the pub" principle.
The problem is, measuring power "at the wheels" isn't any good either because the figures are very sensitive to tyres, tyre pressure, weight on the rollers, etc. OK if you're comparing runs from the same car on the same day but not between vehicles.
I think rolling roads have their uses in tuning, especially when making comparisons between settings that the seat of the pants isn't sensitive enough to notice or when you need to be kicking out 200 odd BHP and tweaking a parameter at the same time. Having mapped a car solely on the road I wouldn't complain if someone gave me a rolling road. There's only so long you can spend at 200 BHP on the A31! Then again, map a car on the rollers and you'll need to finish it off on the road.
Cold morning are good, especially if you have a turbocharged car!
In a Turbo you've got the double whammy of cold intake air to the compressor and an intercooler / chargecooler working really well too!
Kevin