I don't think so vitalidze. If scientists know what performance is available then they will taylor the project to make best use of it. If you give them 1'000 more crunching capability, no problem to make the project need it.
Fundamentally there is no limit. Certain problems will need more power ad infinitum. The trick is to define the problem and the question asked so that it can be computed and analyzed in an acceptable period of time in regard of the available crunching power.

If you take meteorological weather analysis. The most powerful supercomputers do that. And if you see mesh size for simulations we talk many kilometers. This is in inhabited regions. On the oceans it is much worse. Just bring it down to meters in 3 dimensions across all the planet and say the first 30 km of atmosphere and even a quantum supercomputer will not be enough by far. The other element is also the complexity and number of interaction that you will take into account to be computed. The mesh maybe gigantic but at each point you also can compute much more variables. This is why these problems which are similar in terms of complexity and interaction to molecular simulations will always ask for more, more, more, more.......
But they have us here who will give them more, more....

Btw vitalidze I have to stop here I have to add another 12 core cpu.

No matter how fast you can return the data if no one looks at it is worthless! So sometimes just crunching faster is not helpful nor the answer. At some point PEOPLE need to look at the data and if someone can only spend 2 hours per week doing that, then a bazillion units being returned is not better!!

Some GPU projects run out of tasks, space, and do not have the research time to analyze the data in time to produce more tasks to keep the feeders supplied.
Only with massive funding, support and scientific throughput could tens (potentially hundreds) of thousands of systems continuously run GPU tasks.