Of course, current computers are nowhere near capable to try out substantial portion of all 256 bit keys. It requires massive amount of energy and time to perform single bit operation, much larger than kT for room temperature T (several orders of magnitude). Not mentioning costs of trying out the resulting bit sequence for validity as a key. That should be the argument for "256 bit keys are enough".
There isn't any fundamental energy cost imposed by physics here, however. Both because 1) bit flipping can be done in a logically reversible way, just go systematically from 0 to 2^256 - 1 so Landauer's assumptions do not even apply 2) Landauer's idea has been criticized for being vague/badly reasoned. Most weirdly, Landauer assumes that erasure of a bit register in general requires that thermodynamic entropy kln 2 per bit is acquired by the environment. It seems people are confused and can't distinguish information entropy and thermodynamic entropy here. In real computers, erasure of bit register decreases information entropy by kln 2 and increases thermodynamic entropy (by HW-specific amount) associated with the register. These are two different kinds of entropies.
In short, real world energy costs are far higher than Landauer's limit due to current tech limitations, and possible energy cost savings in the future aren't hard limited by Landauer's limit at all. Landauer's idea is simply too problematic. Don't rely on it for any argument about real world.
Finally, don't learn physics from computer science guys, even if their name is Bruce Schneier. Just as you wouldn't learn computer science from physics experts.
There isn't any fundamental energy cost imposed by physics here, however. Both because 1) bit flipping can be done in a logically reversible way, just go systematically from 0 to 2^256 - 1 so Landauer's assumptions do not even apply 2) Landauer's idea has been criticized for being vague/badly reasoned. Most weirdly, Landauer assumes that erasure of a bit register in general requires that thermodynamic entropy kln 2 per bit is acquired by the environment. It seems people are confused and can't distinguish information entropy and thermodynamic entropy here. In real computers, erasure of bit register decreases information entropy by kln 2 and increases thermodynamic entropy (by HW-specific amount) associated with the register. These are two different kinds of entropies.
In short, real world energy costs are far higher than Landauer's limit due to current tech limitations, and possible energy cost savings in the future aren't hard limited by Landauer's limit at all. Landauer's idea is simply too problematic. Don't rely on it for any argument about real world.
Finally, don't learn physics from computer science guys, even if their name is Bruce Schneier. Just as you wouldn't learn computer science from physics experts.