+1 is for file access, since the cpu has that little part of total time (partially) free

Actually this calculation is not correct. Assuming we have n cores (n > 2) we can do the following:
subtract 1 for the system, 1 for all the user processes together like X, IDE, browsers and others and safely multiply by two the number of remaining cores. This will give you approximately the number of parallel builds (the number to -j flag). You can even use that number as a lower limit and you can increase it, but usually higher values will not give you faster builds - but also it will not slow down anything, therefore it is safe (up to a reasonable limit).

I have a FX6300 and the limit is -j7 for me, anything above this and using the rest of the system can be painful

couple things
the kernel is threaded by design
if there are 2 processes running on one core, bout will get 50% of it
CFS group scheduling patch that landed a while ago treats a group of processes as one process
so even if you put -j100 the desktop will be responsive (provided you have enough memory to do that)

also there is nice

but yes, putting more then you have cores gives usually no gain
if i remember good, ncores+1 gave me 1% faster compile then ncores

thats weird
i compiled fine on my old single core
and now i can run things on like 15 threads np

if by painful you mean not snappy as it usually is, try nice