What you are saying is not strictly true, but the point is minor enough that it doesn't really change things one way or the other.
If it is not accurate, let me know how it is innacurate, don't leave me guessing. I like to be as accurate as possible.
More importantly, DNA doesn't "do" anything, merely providing a template.
That is a terrible analogy for what DNA does. Sure, DNA needs the rest of the cell to accomplish its work and even the entire organism, but to call it a template is like calling a program a bit of nonsense in a computer. Like a program without which a computer is just a piece of junk, without DNA a body would be just food for scavengers. Like a program, DNA is information, essential information for the human body, just as a program is essential information to make a computer work.
Building proteins off that template is an extremely biased system
If you mean by the above that it takes a lot of fiddling to get the protein to come out correct, you would be right, however it is DNA itself that sets out how it is to be fiddled with with stop codons, homeoboxes, specific RNA's for specific genes, sets up a control system to tell how much protein is to be produced and when, and much more. So you are totally degrading the tremendous job which DNA does in the organism.
we expend a fair portion of our supercomputing power today figuring out what protein conformations are probable under certain circumstances and which aren't.
If you are trying to make a better protein than nature or to modify it in any way you certainly will need a lot of work to accomplish it which like the rest of your post pretty much verifies what Alamo-Girl's sources have been saying - that it is virtually impossible to create a single functional gene at random. Such a miracle occurring once would be possible though extremely unlikely. However to propose that such a miracle could have occurred not just once but millions and millions of times with the numerous species we have on earth is certainly impossible.
That isn't really a valid analysis -- apples and oranges. You completely missed what I said, so I'll try again. "Difficult to compute" and "improbable" are completely orthogonal to each other. I can trivially compute molecular interactions that are virtually impossible in a real molecular system (as shown by the computation). Computing a particular conformation has the same cost no matter how probable or improbable its actual occurence is.
What we are trying to do (and straining our computational abilities as we do it) is compute the probabilities of an entire molecular phase space. Not just what happens in a specific instance, but what could happen under what conditions and the probability pertaining thereto. The easy case is taking a particular starting point and seeing what the outcome is. Even worse, we often attempt do an inverse computation i.e. given a certain protein outcome, what are the possible starting points that would give that result. Quite frankly, our computers are pretty taxed doing the forward computation, and doing the inverse computation is largely beyond our computational abilities. It is not a symmetric computation, in the same way factoring large composites is vastly more difficult than multiplying the primes that make up the composite.
Biochemistry is computationally probable for the most part, and we can compute specific results with relative ease. Computing the inverse case so that we can manipulate protein systems at will is nigh intractable. Nature just follows the probable pathways. Computing what probable pathways can get you to a specific endpoint is extraordinarily difficult no matter how "common" and probable the protein interaction is. The extremely difficult inverse computation is important because it allows us to thoroughly explore biochemistry (both the probable and improbable), and despite the computational expense, it is often cheaper to do the modeling on computers than actually testing and sifting the astronomical number of permutations in a lab.
In short: difficult to compute is utterly unrelated to probability. We aren't just analyzing what happens from a specific known starting point, we are reverse engineering the entire phase space of possible starting points and possible end points. A vastly different problem, that.