You can do both if you make multidimensional tech areas.
Have the normal "uninspired" tech area produce a reasonable start, but have diminishing returns.
Add a new tech, the "breakthroughs" area.
At each higher level of the breakthroughs area,
you will open another dimension of research into the original tech.
Your multidimensional tech could be something like the following:
code:
| uninspired
| 1 2 3 4 5 6 7 8
---+-------------------------------
B 0| 5 10 15 20 25 30 35 40
T 1| 10 15 20 25 30 35 40 45
# 2| 15 20 25 30 35 40 45 50
1 3| 20 25 30 35 40 45 50 55
The uninspired research will provide hefty practical gains, but soon becomes expensive to add levels too.
A breakthrough will allow you to start working on a new dimension to the problem, which happens to be a tech level that is "fresh", and not yet suffering from the diminishing returns.