If I'm not mistaken, doesn't USB have a way for devices to access the host's memory via DMA? If so, does that mean it's possible for a 'hacked' keyboard to use DMA to write an exploit into the host machine's memory?
Why would you want to use PS3s for a homebrew supercomputing cluster if it means you have to write and optimize code for the SPEs to get benefit out of it? The PS3's linux environment doesn't let you utilize the GPU or all of the built-in SPEs and it doesn't have a lot of RAM available either. It seems like it would be cheaper to build a cluster out of commodity PC parts, and maybe use GPUs+CUDA to get more muscle without having to completely hand-roll your own accelerated computation code (since CUDA is roughly C). I can't imagine that the PS3 would end up cheaper for these purposes, considering it includes a Blu-Ray player along with a bunch of other things you're not going to be using.
I'll enumerate the primary reasons quickly, since I don't expect you to be intimately familiar with the relationship between graphics programmers and graphics driver developers (it's drastically different from Intel's relationship with the X developers):
1) Intel graphics drivers are possibly the most inconsistent drivers on the market. Any given user with a particular Intel chipset might have one of a hundred different driver configurations, as a result of the fact that the chips are bundled with different motherboards which then come with their own driver package... and when you add pre-built machine vendors into the mix the situation is only worse. If their driver quality was extremely high across the board, this wouldn't be an issue, but...
2) Intel graphics drivers have a bad stability track record, at least on Windows. They have a tendency to return invalid/nonsensical error codes from driver calls that shouldn't be able to fail, or to silently fail out inside a driver call instead of returning the error code they're supposed to... resulting in graphics programmers having to special-case handling of individual Intel graphics chipsets (and even driver revisions). In my case, I ended up just having to shut off entire blocks of my hardware-accelerated pipeline on Intel chipsets and replace them with custom software implementations to avoid the incredible hassle involved in coming up with specific fixes. (The wide variety of chipsets and drivers out there meant that for my particular project - an indie game - it was impossible to ensure that I had worked around every bug a user was likely to hit, so I had to just opt out of hardware accel in problem areas entirely).
3) Intel graphics chipsets have sub-par performance across the board, despite marketing claims otherwise. This is mostly problematic for people developing 'cutting-edge' games software, where it creates a 'he-said-she-said' situation with a game developer/publisher claiming that a user's video chipset is insufficient to run a game while Intel claims the complete opposite. (in most cases, Intel is lying.) This is particularly troublesome in areas like support for cutting-edge shader technology, where an Intel chipset may 'support' a feature like Pixel Shader Model 3.0 but implement it in such a way to make it completely unusable. Users don't benefit from this, and neither do developers.
4) Intel graphics chipsets harm the add-on graphics market by discouraging users from picking up a (significantly better) bargain video card from NVidia/ATI for $50 and dropping it into their machine. This hurts everyone because even though that bargain card is significantly better (and most likely more reliable), the user already 'paid' for the integrated chipset on their motherboard, and the documentation that comes with it attempts to make them believe that they don't need a video card. I consider this a dramatic step backward compared to the situation years ago, when integrated graphics chipsets were unheard of and people instead had the option of 'bargain 2d' video cards like Trident or Matrox that would do everything needed for desktop 2D, but also had the option of fairly affordable 3D accelerator cards if they wanted to play games occasionally.
On the bright side, most integrated ATI/NVidia GPUs these days are mature enough to be able to run games acceptably and meet the needs of a typical user. The only thing really holding the market back here, in my opinion, is Intel's insistence on marketing inferior products instead of partnering with ATI or NVidia to please their customers.
Of course, this is unrelated to your point that their Linux/Free Software support is superb, as is their documentation - I'm inclined to agree with you here, but it unfortunately doesn't do much to outweigh their other grievous sins.