Frank,
Well, there are some possibilites for flaw.
X3 is basically a CMOS type of sensor. The most basic problem of these sensors is not RGB separation and such, but pixel-to-pixel uniformity. They utilize transistors in each pixel. These transistors have their parameters a little vary on the pixel to pixel basis. These variations produce picture noise, if not corrected. Actually, in high-end cameras they are digitally corrected, which is pretty expensive (like Canon D30). In cheap cameras they are just ignored, resulting in noisy picture.
Mid-range cameras, like f707 and such, are using CCDs which do no not suffer from this problem.
Now, Foveon only mention they somehow improved the uniformity, but do not debrief by how much and how they did it. Note, this is the real problem of CMOS sensors, not a color division, which can be made by conventional methods. I'd like to hear from Foveon more details on this, instead big buzz on color division.
Second, to separate colors better they should subtract signals from all three layers, which increases noise to some degree. This might limit their high-ISO abilities.
Other question is a quantum efficiency, or sensor's ability to convert light to electrons efficiently. This depends on how they create doping profiles so photoelectones could be effectively collected. A generic CMOS process obviously does not have it's doping profiles optimized for x3 sensor. Foveon might take it as is, or optimize, where optimization means higher cost.
Also, a generic CMOS process often has relatively high leakage by CCD standards. This might kill x3 low light sensitivity. Alternatively, Foveon might optimize process for leakage, which could cost more.
Further, while Foveon claims they use a CMOS process, they actually use it's more rare flavor, called triple-well CMOS.
This flavor is commonly used for mixed-signal integration or for BiCMOS logic. This flavour is a little more expensive than a generic CMOS process and somewhat less widely available. Morever, they require a lightly-doped substrate version of this process. On the other hand, there are probably 10 to 20 companies having the this version in place, so it should not be a major obsatcle.
All in all, at this stage it's hard for me to say if this is a jewel or junk. I'd wait to see a finished product and independent reliable tests.
--Vladimir.