http://arstechnica.com/news.ars/post/20080702-intel-an-expensive-many-core-future-is-ahead-of-us.html

Expensive for who?  For software developers of course.  It seems to me that the silicon industry is trying very hard to obscure a basic fact: even if new technologies such as Software Transactional Memory become common, even if you give us 100 cores to play with, even if things like the Parallel Extensions from Microsoft are elegant and easy to use, even if Intel's compiler research turns out to be a huge help,  not all problems can benefit from parallelism.  In many cases, programmers can go against what would today be considered good practices and make copies of huge shared data structures, (at least we're not being told that we're never going to have more than 4GB of memory) in order to reduce data sharing between threads.  However, there are many problems that need shared read-write data.  Throwing massive numbers of cores at these problems will result in performance slower than single-core performance as resources are eaten up acquiring locks.  On the Windows platform, all of our GUI technologies still use a "compartment" model whereby objects are owned by a single very special thread and we are not alowed to touch them except by marshaling onto this Special message pump.  What good are these 100 core systems going to do my WPF applications?