That's not entirely correct. Most chassis dynos do measure torque and calculate hp, but there are some dynos like the DynoJet that actually measure power and calculate torque using RPM data. Remember that power is a measure of work over time. Dynojets work purely off the weight and inertia of their large, heavy rollers. By measuring how quickly those rollers are accelerated over time, they can then derive a direct power measurement. Then the software uses the RPM data to calculate back torque. You'll notice, for example, that you can actually do a pull on a Dynojet without using an RPM pickup and it will only plot power over time because it does not have RPM data to use for its torque calculation.

Road dyno software works on a similar principle. It measures the car's actual on-road acceleration, and then using parameters like weight, frontal area, and drag coefficient, the software calculates the work done over time by the engine and then can derive power and torque just like a chassis dyno. Given optimal conditions (100% flat road, no wind, extremely high data capture rate, etc), the power/torque representation from road dyno software would be exactly as accurate as the power/torque derived by a chassis dyno. Of course we don't live in a perfect world. I've found that the main limitation in most road dyno software or measurements is the data rate of whatever logging software or interface being used to acquire the data. The lower data rate makes it more difficult for the software to filter out noise while still keeping a good level of detail in the dyno plot. This is where chassis dynos shine because they generally have a very high data capture rate and are able to show greater detail in a lot of cases.

Of course chassis dynos can have their drawbacks as well. Using the OP as a perfect example, his car would only make 23psi on the chassis dyno, but as soon as he put it on the street, it makes 26psi easily. This means that either the dyno was not providing as much load as his car experiences on a real road, or the car was not getting as much airflow into the intake/intercooler/radiator on the chassis dyno as it does under real world conditions. Most likely its a combination of several factors. Another issue to consider is the air quality inside the dyno facility. A lot of chassis dynos are set up in a manner that causes the dyno room to fill up with exhaust fumes over time and these fumes are then sucked back into the engine through the intake. Obviously exhaust fumes don't contain nearly as much O2 as open air, so this can greatly throw off a/f ratio measurements as well as power levels.

Its always important to remember that these are tools we use and they all have their benefits and drawbacks. Exploring those differences is important and will hopefully help us optimize our tools in the long run.

Thanks
-- Ed