I think it would be a great idea just to see how much the HP changes when you don't change out silencer packing or clean the air filter.
Dynos are a great tool to compare other things with, not so great at determining how fast a bike is.
Best thing to do is make that curve look smooth and wide, not jagged and peaky.
For example, if a new set of reeds drops your peak 1-2HP lower but fills in several valleys and picks up HP in areas other than peak, making the graph look more like a curve than a cliff, I'd call that new set of reeds permanent.
If you're riding on dirt tracks and you're constantly modulating the throttle to manage wheel spin, wouldn't you rather have a consistent, smooth powerband to work with instead of a ragged and sharp powerband with a few extra HP up top?
A dyno is great for tuning of that sort...the numbers by themselves are arbitrary, but the curve will reflect the powerband itself quite accurately.
You can take a mean average of the distance between the curve and baseline and see HP reflected in overall RPM ranges.
Imagine the volume that a steady curve has compared to a peaky curve like a banshee has...even if the banshee has an extra 20HP, you've got consistent power throughout the shift range.
So when the banshee falls flat on its face after a shift, you're still tooling along with almost the same power as where your peak HP is.
HP over a broad RPM range is much preferred to lots of HP in a narrow range.

Effectively, that's what I think a dyno is useful for, making consistent power instead of absolute power.
That, and seeing what kind of improvements little things can do for an engine.