I agree with everything you are saying. What I was trying to say in my original post is that alot of the lock-in that Windows used to have is slowly going away. They do, however, have alot of mindshare which keeps them the "default" action for alot of users, just like people type in Google. And you are absolutely right, having a 10% better search engine or a slightly cheaper OS is not going to take away mindshare.

Having said that, it isn't entirely clear to me that most users "choose" Windows. Often vendors or "IT support" choose their OS for them. For example, my wife moved to a Mac on my recommendation and it took her about a week to get up to speed, and she is not a tech person. The reason the transition was mostly seamless, though, is because she was using the web for most of her day-to-day computing, like email, facebook, etc.

Once all the barriers are gone, people won't switch instantly, but system vendors might push the switch to reduce costs. Alternatively, I could see the cellphone evolving into a platform that people use over Windows and displacing Windows and PCs entirely over the next 10 years. Only time will tell. Microsoft is incredibly resourceful, and their efforts with Bing, especially in terms of mindshare and advertising is impressive. They've actually made headway in gaining search market share, something that many thought couldn't happen, even though it is only a small gain.

Indeed. Though Cory Doctorow has a piece at the Guardian (http://www.guardian.co.uk/technology/2009/sep/02/... that talks about the downside to the cloud. He correctly points out that it could end up costing users more and lock them in more strongly. Fortunately, companies like Google are, for now, allowing an easy path for migration out of their services. The question is how long will this last?

I'm not sure how the difficulty scales with processor count. For example, for some code I am writing I added 1 thread and things got *much* harder. The code works now, but (tooting my own horn here) I don't think Joe Programmer would have been able to fix some of the bugs I had to deal with in any reasonable amount of time.

The challenge as core count scales is getting performance out of them. I'm not sure how many cores you need to get to before this is as hard as just writing a correct parallel program. Addressing your point 2 (which my research group and startup have/are working on technologies to address) definitely makes performance scaling easier.

As for ASIPs, that assumes that the ASIP doesn't get it's performance through multicore like Cisco's quantum flow or some other network processors. If it does, then you are back to square one. If there is a lot of specialized hardware that is hard to program (like Intel's IXP) you might even be at square -1. I think the real win for ASIPs is when:
1) You care about power (which is almost always now)
2) You can build something in the ASIP that lets its energy-delay product out pace a general purpose multicore enough to get people to buy it and program it.
3) The target market is large enough to support the NREs and mask costs at the price people are willing to pay for the ASIP's value add.

Fortunately for ASIP folks I think there are plenty of markets where this is true. At least until mask costs and design complexity become untenable. ;)