I read quite a few threads talking about how the newest MPs could/would/might be faster with single-threaded apps after Grand Central was implemented. I'm always up for learning if you can provide some info.
Well, first, you can't make a non-parallel app parallel at runtime, as a practical matter.
The only way an app would be sped up is because the libraries it uses are able to run faster. An OpenGL app might get faster if the graphics card gets faster, or the libraries get faster, or the libraries become multithreaded. You might recall this happened in 10.4.6 or thereabouts to OpenGL on Intel. (PPC got the same gain in 10.5.) Apple threaded their OpenGL code, and even a single-thread app that was calling those libraries was able to experience the gain, however, I do think the app technically had to request multi-threading in OpenGL to guarantee it wouldn't cause problems. The demo app was World of Warcraft. Frame rate basically doubled when you clicked the switch.
Another example is the Accelerate API that Apple introduced in 10.3 ( I think). If the programmer used the Accelerate API, their program would automatically make use of AltiVec, if it existed. Later, on Intel, it would automatically make use of SSE, if it existed. If it didn't exist, Accelerate would just use normal CPU instructions.
So, only libraries that are being called can possibly benefit from Grand Central without the app being rewritten. And there aren't really any libraries except the well-known ones above that are likely to see those kind of gains. And if they see those gains, there was no reason why they needed to wait for Grand Central.
Grand Central is generally understood to be like OpenMP (a compiler and runtime parallelization system), while OpenCL, which is somewhat like the Accelerate API, and will allow GPUs to be used to calculations by general apps.