Every dyno is going to read things differently. This is inherent due to several factors, which include, but are not limited to, atmospheric conditions (temp, barometric readings, etc), testing conditions (do they place a fan in front of the radiator, how tight did they tie down the vehicle, etc), type of dyno (eddy current or acceleration), or manufacturer of the dyno itself (Mustang, Dynojet, Superflow, etc.). Even the quality of the gas can have different results. Each is going to provide different readings.
It is a well known that different brands of dynos produce different results. There is a big rift in the hotrod community as to whether the Dynojet or Mustang dyno is better. Each provides a number, but they are usually different.
With a dyno there is one major thing you need to consider, that being am I being consistent? As stated, there are two major reasons to use a dyno in the first place.
First is tuning. A dyno will allow your tuner get a better than baseline tune (doing a road tune will provide a better tune as these produce real world driving conditions). Any type of dyno will work in this situation because you are only trying to get the most out of the vehicle no matter the power/torque levels.
Secondly is to observe a difference. Before you do work on a vehicle of a performance nature, it's a good habit to get into to put the vehicle on a dyno and test it out to see what are it's current power/torque levels. In doing so, you give yourself a baseline. When you do the modification, take the vehicle back to the same dyno and test it again. This will give you the difference. Why use the same dyno? Simply to allow the vehicle a fair shake and to be as accurate against the baseline run as possible. No two dyno runs are going to be completely accurate against each other. If you are using two different dynos to check your results, these numbers are going to be farther away from the truth of the gain (or decrease). You want as close as you can get and using the same dyno is about the only way to do it.
I don't have any hard and fast numbers for you as to how close is a dyno really. It mainly depends on how often the dyno is calibrated and how well that calibration is done.
One thing of note here is that most dynos take into account the atmospheric conditions encountered at the time of the dyno run. The computer will usually adjust the figures based on these and spit out an answer which is the adjusted value at mean sea level under perfect conditions. This allows the vehicle owner to have as close to standard of a number as which it can be. They can then take the number and compare it to other dyno runs and know the numbers are about as close as can be expected.