"simply because my best guess was that I am the only person who will ever want feature x"
You may have been underestimating 'the others'... "Release early, release often" means release it, even if you think it's (still) useless junk. Just label it as that, and perhaps others will find it better than useless junk, or if needed maybe clean it up and turn it into something you never even thought it could be.
At least send a message 'listen guys, this is what I threw together for myself and here is why', or put up a webpage on a blog or wiki somewhere with your patches and mention the site once on the mailing list.
Do it for the others, whom may surprise you.
A lot of programmers with good intentions end up never releasing what they've made, and what could have turned into something great, just because they want to 'clean it up first', or because they think 'nobody would want it' (they wanted it, so somebody did, making it less than unlikely that somebody else wants it too). Release it, just be honest and say that even you the creator thinks it's dirty and useless. Perhaps others disagree about the 'useless', or are better/faster than you in cleaning it up, or maybe it inspires others to make something similar, or more advanced, in 'the right way'.
"Mike wonders why the kernel tries so hard, rather than just failing allocation requests when memory gets too tight."
I realize this is formulated in a negative way, with no prior reservation of resources, but erm, it was fast and easy right now and gave a sufficient response to the thread with the lowest possible latency, and if and when it ever becomes important I'll reformulate it nicely right before it's needed, and until that time those resources stay available for other uses. So be warned, here it comes: Probably that is because Mike doesn't know what lazy allocation means, why it is used, and that that means that there is not an allocation request to fail when the OOM condition happens?
Hmm, I sound so arrogant in this post that I'm probably wrong... But I can't help but feel that I'm pretty close to being right...
"If it's not superscalar, why does it need a branch predictor?"
Because it has a pipeline and without a predictor a branch means a pipeline stall until the branch comes out the execution stage. With a predictor, even a simple one, it means a pipeline flush only when the predictor is wrong, which means you can gain a lot of cycles with even a very simple predictor.
And slow down interactive response such as seeking?
No, latency is _not_ good.
"Gotcha, you snot-necked weenies!" -- Post Bros. Comics