[gdavisson]

The thermodynamic limit cited (Landauer's cost of erasure) to show that 128 bits is enough (barring algorithmic advances) doesn't strictly apply. Charles Bennett[1] pointed out that any irreversible computation (i.e. one that involves erasure) can be made reversible by saving all the intermediate results (rather than erasing them), printing the result, then running the computation backward and letting it eat up all of the saved intermediate results. While this approach to computation isn't directly applicable here, it does show that you can't count on the thermodymanic cost to keep you safe.

[1] C. H. Bennett, "Logical Reversibility of Computation" _IBM Journal of Research and Development_ 17:525-532 (November, 1973), http://www.cs.princeton.edu/courses/archive/fall04/cos576/pa...

[pbsd]

There is another more fundamental limit, credited to Margolus and Levitin [1], which puts the maximum speed of a (quantum) gate at 2E / h operations per second for a system of energy E and Planck constant h. Given the energy of the sun quoted in the article, this puts the upper limit of bit flips per second at 2^193. Seth Lloyd's 'Ultimate physical limits to computation' [2] goes into further detail on this subject.

However, the notion of zero energy dissipation (scalable quantum computers will likely need error correction, which implies bit erasure) on any real system is ludicrous, and I think the Landauer bound models reality better than Margolus-Levitin.

[1] http://arxiv.org/abs/quant-ph/9710043

[2] http://arxiv.org/abs/quant-ph/9908043

[SAI_Peregrinus]

Either way the limits are so far out from the computational capability of humans in the forseeable future (next few hundred years) that they're both massive overkill. When you really NEED that much security you have to worry about the loyalty of your nation's military and how good all those armed guards will be at keeping you from being waterboarded.