I must admit I used to be one of those goofballs who tried to eke out every single possible performance gain. But it only makes sense on underpowered systems.
When it comes to graphics features- including things like Windows 7's Aero Glass- If they are hardware accelerated, you typically won't see a performance improvement by disabling them. Sometimes you will even see a performance loss, this is particularly the case when it comes to desktop compositing, which as Camerongray notes, offloads a lot of the processing to the GPU, which would otherwise be mostly idle.
You can argue that some features "serves no purpose to me other than aestethics and even if it doesn't use much resources it's still a drain and pointless to keep on." but at some point you need to quantify it, which would mean actual testing. And I've yet to see any actual testing with any semi-modern system that shows any particular performance gain from disabling the sorts of features you mention.
I think the propensity and popularity of the idea of "secret tweaks" is due to the proliferation of the mindset of folks like the guy who talks about disabling services (BlackViper?). Instead of relying on real-world metrics and measurements, there is a reliance on perception- "it feels about twice as fast now" And that get's proliferated and cited as a "fact" elsewhere, leading others to try it, and because they were told it would be faster they delude themselves into "feeling" that it is faster.
The best example of this, I think, is the old "myth" that Windows XP had Superfetch and you could enable it with a registry tweak. Basically, you could add a specific key to the registry from Vista, and XP would magically upgrade it's "prefetch" feature to a full superfetch implementation. Despite the fact that this is complete nonsense, people ate it up- you can look around on forums from the time and find people swearing up and down that their system is made much faster by changing this registry key- a registry key which does absolutely nothing and isn't used by any component of Windows XP at all.
We can see, once again, the same story for the old "DirectX 10 on Windows XP"- look around on forums- including this one, and certain users swear up and down that after installing that special "DX10 on XP" patch, all their programs and games are running much faster. Look further on this forum and you might even find where I investigated the claim and found that the "DX10 on XP" were nothing more than a non-working wrapper that delegated DX10 calls to DX9, meaning that it was impossible for the change to improve performance since it didn't actually add any new mechanic.