I think an engine dyno is much better. Think about it. On an engine dyno you have isolated the engine from the vehicle. This makes it easier to make the engine make more power. There are less extraneous variables to cloud the numbers.
As installed in a vehicle and run on a chassis dyno you have lots of variables that have nothing to do with the engine whatsoever. Tire pressure, rotating mass after the flywheel, torque converter (if an auto), correction factors, etc, etc.
On an engine dyno you basically have the actual torque and hp an engine makes, and a correction factor. That's it. That makes it easier to see small changes in power from the changes you make. I would say a good engine dyno will be within about 1% from run to run. That means that if you have a 500 hp engine, each dyno run should be within 5 hp. This makes it relatively easy to see the small effects of tuning changes. On a chassis dyno there are so many other variables that a small positive tuning change could easily be masked by some other variable changing at the same time.
The ultimate engine dyno is the track. You can't argue with the final result. The second best engine dyno is an "engine dyno". You can make changes and see the results faster than track results. The third best engine dyno is a chassis dyno. It can be useful, it's just the least useful of the three, in my opinion.
My recommendation is to use an engine dyno first. Then fine tune at the track. I picked up over 30 hp on my last engine by tuning it on an engine dyno, just from carb jetting, ignition timing, and carb spacers.
JMO,
paulie