I've spent a lot of time, and I do mean a lot of time, thinking about computerizing character generators and even the entire D&D rule set. It's non-trivial. I'm fairly confident that given enough free time (or paid time), I could come up with something cool, but I hesitate to say that it would sell well to a wide audience. Everything from here on out is related to how I, as a software engineer, think that a computerized character-generator should work, meaning IMO.
In order to support everyone, you need to make it easy for GMs to enter their arbitrarily complex house rules. I don't use character generators because inevitably I come to a point where I'm using a supplement, or a house rule, that the generator doesn't support. At that point it's, Wham, bam, thanks for playing because the entire point of the generator (mainly validation), is out the window. That means that logistically you need to be able to support supplements as soon as they ship (assuming that you want to support said supplements quickly), meaning that you need to coordinate print and electronic releases. In my view, this gives you one of two development models: (1) you have a simple document editor that basically just sums up numbers in boxes but doesn't really validate anything (see Mad Irishman's form-fillable PDFs) or (2) you have a highly abstract system, tied together loosely by some sort of scripting language, and you keep a permanent staff of developers to code each supplement.
In reality, I think that Mark is at least partially correct. From a business standpoint, if you're going to produce an electronic character generator, the electronic generator will, on some level, influence the published material.