aLoha!
i was thinking today and one of thoughts was that for new processors with 4,8 cores and more scheduler and gcc can work different.
i'm not really sure how this is done right now, but i guess that it is a concurrent model - when one thread works separately of others, so they are like in async mode and when one thread is ended up it starts the other, i see that on processor load, that most of my programs are using one kernel, sometimes 2, but really it is not the problem of building program, it is about how the kernel architecture works.
and i think that now this can be done not in parallel or async mode but simultaneous, so when process starts kernel schedules work not for first thread, then second etc - it loads program into memory analyses cross processes and starts simultaneous work, in the end it will save the process map, so on next start it will not analyse that process again, but will use the map or maybe that map while loading can be a kind of memory image, so it will load the end state of the process without recompiling it again.
i think i've explained the idea, i'm not so much familiar with this kitchen, so i can probably add something later, and if you will like it maybe we will have some discussion prob.
have fun ,~,