I have also only compiled the kernel once.. a long long time ago when 2.6 had recently been released, and i was trying to turn on the hyper-threading option.
Originally Posted by hkothari
I'm assuming one advantage is it runs faster because it's customized for your hardware
Not necessarily. it's all relative
Basically the main advantage is to preload (or prevent preload) modules which you may need. besides that, there are tons of options you can turn on or off depending on your need. so, if you were to preload a lot of modules or turn on some options that aren't on by default, you might actually make it slower.
how fast or how slow depends on the options and modules you select. also, if someone was to say it will shave off 10 microseconds per instruction, for instance, to a desktop user like myself that wouldn't mean much, but it would really be a huge deal for people using it in real-time envs. and simulations, etc.
i think no one can really quantify how fast or slow it would be unless they had in-depth knowledge of the hardware and the options you want to select.
you might have to run some stress tests to get exact figures.
in conclusion, for lay-people like me, the pre-built kernels are far too efficient already, and the time and effort required to compile it is just not worth it. But, you know your requirements best.