First, you can see that the current Pentium D really is almost exactly twice the size of a Pentium 4, and the two cores are arranged side by side. The 90nm Pentium D is also really frickin' big.
Once the Pentium 4/D architecture makes the move to 65nm with the Cedar Mill and Presler parts, also based on the Netburst microarchitecture, the die size gets smaller. You can see how the dual-core Presler really is two Cedar Mill Pentium 4 chips situated together on the same package. Because the Pentium D's two cores communicate with one another over the front-side bus and not via any sort of on-chip data path, the move to Presler's package-based arrangement makes some sense. Each of Presler's two cores can be cut from anywhere in any wafer, and a defect in a single core won't torpedo the adjacent core. This arrangement should lead to higher yields and less manufacturing risk for Intel.
Finally, at the top, we have Conroe, the CPU based on Intel's next-generation microarchitecture just announced today. Notice that Conroe is, shockingly, ready to drop into an LGA775 socket just like any current Pentium 4 or D. I don't have any definitive word on what specific chipset or motherboard changes will be required to support Conroe, but the fact that the physical socket remains the same represents somewhat unusual continuity from Intel on that front. The use of the LGA775 socket should confirm for you another detail about the new microarchitecture: it will indeed use the same front-side bus as current Pentium processors, although perhaps at a higher speed. I asked for details on the bus speed for Conroe, but the Intel folks weren't willing to say.
Also note that Conroe's die size appears to be smaller, all told, than the total die area of the two Netburst cores on Presler. Both chips are made with the same 65nm fab process, so the comparison is fairly direct. Presler's dual-chip-per-package arrangement could very well make it cheaper to manufacture than Conroe, however.