On the one hand, the point of having a steam-powered computer is to have a steam-powered computer. What fun is a virtual steam-powered computer? You can't scald anyone with that!
On the other hand, yes, in the year 2010 you really should master the art of building the virtual Analytical Engine before you go for the real thing. One reason, I believe, is that the current draft of the design of the Analytical Engine is more akin to the world's largest collection of back-of-the-envelope sketches than to a finished blueprint. There was never one canonical design for the Engine, because so little of it was ever actually realized in hardware. Babbage had a lot of fun swapping parts in and out in his imagination.
So there needs to be a bit of each approach. Build the virtual version first, in an environment where building 1024 identical custom-machined ratchets and pawls is just a matter of ten copy-and-pastes. Then a nail down a plan for what you intend to realize in hardware. And then build it. If you can find the money.
There will also have to be some back-and-forth. One of the central lessons of engineering is that it's more than just CAD. It isn't until you try to build a thing that you fully understand its design, if only because it is difficult to imagine the constraints that reality imposes on your design -- and, more embarrassingly, that the earlier parts of your construction impose on the later parts -- until you've actually gotten into the middle of the job. This is one reason why these projects are so expensive: They are, in part, research projects. I believe that the folks who assembled the Difference Engine weren't entirely sure, going in, that the darned thing was actually going to work as well as it does. And they made some crucial tweaks during the assembly process.