This is a vastly important development that probably means several things for the future of semiconductors, and especially for CPUs. Intel must not be confident in its ability to resolve the Prescott core's heat and power problems at 90nm, despite an extensive redesign involving many more pipeline stages and related tweaks. Looks like this gent was right in suggesting semiconductor scaling died at 90nm.
The move away from the Netburst architecture would seem to signify a fundamental shift away from Moore's Law-based assumptions in Intel's processor design philosophy. Netburst's design appears to be largely predicated on the assumption that transistor budgets will grow, die sizes will shrink, and clock speeds will rise as they have in the past.
Of course, Netburst hasn't uniformly produced performance gains as its transistor count has grown over time, and that may be part of the problem, related to but not entirely tied to the heat and power problems at 90nm. Already, rumor has it that future performance gains out of the Pentium M-based cores will likely come in the form of explicit parallelism using dual cores rather than innovation inside the CPU core itself.
If you haven't yet, go what Bob Colwell's recent talk at Stanford to learn more about what the CPU architecture guys are up against. These latest developments certainly suggest his analysis was dead on correct.
Update: Intel has announced the change in plans, citing a desire to accelerate development of dual-core CPU designs. (Thanks Frank.)