There are trade-offs either way you go. With a complete system you are correct, you must wait for all of it to hit steady-state. In closed-loop systems though, this isn't very long. If you've got a reservoir of even a couple liters, it'll add considerable time.
Any data taken under properly controlled conditions is relevent. It's knowing the error bands associated with all your miscellaneous measurement equipment that is almost universally overlooked.
The other point I was making is that while your test conditions would display block performance, end users could still be disappointed because their complete package:
1) Couldn't meet the flowrates that you used (multiple tests over a very wide range address this)
2) Didn't know where their flowrate would fall and find out it's lower than they thought it would be
3) Don't have the ventilation/radiator area/ambient temperature to match the fluid temperature you tested (multiple tests over a wide water temp range address this)
4) Didn't know what their fluid temperature would be based on their radiator setup and it's higher than they thought it would be.
Where I'm going is that you could invest literally hundreds of hours to test a single block over a wide range of realistic flows and temperatures (not to mention power ratings, but I'll let you fill them in as points 5 and 6 in the list
). In spite of all this, folks would have a hard time correlating the data to a completed system.
Sure, they may know that block "A" did better across the board than block "B". But you don't need hundreds of testing hours to determine this. It's an interesting debate topic that has no single right answer.
In the end you'll never satisfy everyone with testing. Some want precisely what you've outlined. Others want to see systems tested. Many (most?) haven't a clue about what "controlled conditions" means (including most reviewers). Discussions such as this one will hopefully change this last fact.