They are as good as the data you put in them. You have to have the correct weight, so a too full/too empty gas tank can skew the results. You have to make sure that the road is level. You have to correctly calibrate for how much your front end rises and your rear end squats on acceleration.
If you get all of that dead on, they are suposed to be nearly as accurate as a good chassis dyno. If you always test at the same place, always fill up before testing and independently confirm the results at a drag strip fairly often, the data should remain very close and give you good back to back results.
So, if you test, change something and test again, you will be able to see fairly small changes, probably down to 3-4 hp.