I know something that has long been missing in the kernel, in my opinion, is support for socket migration implemented purely in the kernel. This is a roadblock to using Linux in an high availability environment w/out either using user-land tools to acheive a "stateful" fail over or using some kind of hardware load balancer, etc.
It's been talked about every once in a while, and it's a ball that someone will pick up and run with once in a while, and then it seems that the ball is inevitably dropped. I'm not sure why though, and I'm far from a kernel developer. A lot of what makes it tick is still voodoo and black magic to me. ;)
That's my $0.02.