What about Comodo's Defender? You can set it up to automatically sandbox any suspicious programs (unsigned for example) and any suspicious behaviours will be denied and reported. Certainly it is not a silver bullet but I have had good experience with it after it detected a malware hidden in my input method program (which wasn't detected by MSE). The developer site was breached and a modified version was uploaded, comodo alerts me that the program was trying to access the internet.

Although I haven't read tor document in depth, I think blocking certain tor hidden services is doable. A tor node with hidden service will 'advertise' it services on randomly chosen nodes (introductions point), those who want to connect to the hidden service choose one random node (rendezvous point), ask those introductions point to relay the message to the hidden service node, which will initiate the connection by connecting to the chosen rendezvous point (extra step of redirection, I know). So if a node owner want to block a hidden services, he only need to blacklist that .onion domain, forbidding it from being advertised on his node.

Pretty good, it is even more stable than Fedora, rarely do things break, and even if they do there will be announcement on how to prevent/fix it. Still, setting up my own DE is a pain, for example I still can't do tethering with my Android phone without some magical configuration.

While it is certainly possible for some cases, IMO the programmer should rather know and change their design to avoid it than relying on compiler optimisation.

You don't need to guess which machine code will it be turned into nor how every transistors work in your CPU while running it, but you will generally need to understand the mechanism of what you are using. For example in C++ don't call virtual method in a tight loop if you want performance, you will need to understand the idea behind virtual method to know that.

Us Asian all have small finger, that helps a bit with the typing, and there are some Android phones with qwerty keyboard too.
Regarding to speed, I agree that Symbian feels much faster, still I hate how they only includea minimal amount of RAM in their phone, my last phone (a Nokia 5230 with 128 Mb RAM, before I dropped it to death) could only open ~3 tabs with Opera Mobile before running out of memory. Nevertheless it was a good phone. But now Nokia has declared Symbian to be a burning ship, I see no reason to use it anymore. That, and with my personal hatred to Nokia for killing off Meego/Meltemi/Qt and then siding with Microsoft make that 5230 the last Nokia phone I buy.

Opera mini does the same, that is why it is so lightweight and can render (albeit sometimes incorrectly) fairly complex webpages on very weak phones, it even re-encode images to webp format to reduce file size. Amazon Silk also does that too, so it is nothing new.
Back to the topic, for the same price for a Asha 303, you could get something like a Samsung Galaxy Pocket, which has GPS and double amount of RAM. In my country (Vietnam, a 3rd world one) Android is gaining market, even at the lower end segment while Nokia is losing out rapidly. I was surprised that if someone I know has proper web-browsing capability now, then it is most likely to be an Android phone or, sigh, iPhone.

Because the driver situation isn't too good, all open source drivers only support up to OpenGL 3.0, performance is somewhat abyssmal, and the stability isn't that good. With my previous nvidia cards, I haven't had one kernel panic related to the graphic card, while with my current integrated Intel card (on my core i5) I have lost count of the time already (suspend/hibernate memory corruption problem). Binary blobs from NVIDIA/AMD aren't good either.
Nonetheless, it is very interesting to know that Valve prefer the open source driver to the closed one http://www.paranormal-entertainment.com/idr/blog/posts/2012-07-19T18%3A54%3A37Z-The_zombies_cometh/ , if so, then I hope Valve entry into Linux gaming will help improving the situation.

Windows UAC prompt runs on a different session (the secured on) instead of the normal session so that hook-based keylogger won't work any more. Also it has higher integrity level than normal apps so keylogger can't inject their version of dlls to steal the password, well, as long as someone doesn't run the keylogger at higher IL anyway.

Well, even if the user is the sole user with admin right, launched process still runs under with lower privilige than full full admin rights, those that need them will be asked by an UAC prompt. And any user who cares about his stuff will always use a password anyway.

I agree, Windows 7 is a very good OS and IMO, regarding to security GNU/Linux can learn a few things from them, like the MAC mechanism preventing keylog from knowing your administrator password (the UAC prompt has higher IL value), with most GNU/Linux distro any apps can sniff your typed root password. Maybe by using SELinux we can prevent this but I haven't seen any distro implements it yet.

I guess their new Metro GUI uses graphical images more extensively, so any kinds of speed up is good, leaving CPU time for other usage or just preventing your CPU to heat up. And browsers such as IE will benefit from this too, as they have to render a lot of images (I don't know about Firefox, maybe they use libjpeg).
Note that some linux distros have already done this by switching to libjpeg-turbo a while ago.

It seems that they use -O3 optimization in GCC, which is dangerous as it could lead to code running incorrectly, even gentoo recommend against it. I think it would be better if they optimize dalvik VM instead of changind build flag.

Er, currently the compositor on X runs independently from the server, a lot of data has to pass between them, creating unneccessary round trips and degrade the peformance (wayland is an effort to unite the server and the compositor). So I don't really think adding compositor will make much performance improvement or any at all.