Great video!  Really good discussion in this one.

One of the things that was touched on in the video was garbage collection / managed code at the kernel level of the system architecture.  This is very exciting in the sense that it validates .NET and makes it something that we programmers perceive as the way to program - if the big dogs are doing it and trust it all the way to the kernel, it's good enough for me.

This isn't going to eliminate the performance problems associated with managed code, though.  In fact, it could exascerbate it.  When talking about these problems, why don't I ever hear people talk about possible hardware solutions to things like garbage collection / memory allocation / etc.  My school is doing research in this area:

http://www.csrl.unt.edu/~kavi/Research/IRAM.html

I realize that this is far off, but the discussion was pretty far into the future as well, so I was just curious why I've never heard anyone mention implementing hardware solutions for these problems.

Although I'm not on the team that is doing this research, I've seen some of the performance improvements they are touting just re-implementing java's GC in hardware and they are impressive.  Is this something Microsoft is investigating?  It seems like if everyone believes in managed code as much as I think they do, then someone should be willing to burn that belief onto silicon.