[embee]

You need to understand the fundamental principles and experimental error factors involved in testing. Inertia dynos without controlled conditions aren't particularly accurate/reliable/repeatable.
In order to determine engine power for certification/homologation purposes it must be done at steady state for a minimum period of time with all the environmental parameters held essentially constant. That means absorption dynamometers with temperature control for a start. These are expensive/complex, hence the use of inertia dynos by most "shops".
Running in different gears operates the dyno over different speed ranges for a start, never mind any effects within the power unit. Taking results from the dyno using different "road" speeds isn't going to give necessarily particularly comparable results.
How repeatable/accurate is the dyno anyway? It relies on measuring the acceleration of the rollers and calculating the input power based on the inertia and instantaneous speed. The bike is accelerating the wheel as well as the dyno, if the dyno calcs don't take the wheel into account the effect of being in different gears will change the result. The rolling resistance will also change with speed for both the dyno rollers themselves and the tyre/roller contact.
Lots of factors.
Inertia dynos are useful for assessment but +/-3% uncertainty is probably fairly good going, +/-5% wouldn't be unusual.