This is not as comfortable a margin as it seems. If a direct implementation of the algorithm takes a few million qubits, it may well be that both better algorit

This is not as comfortable a margin as it seems. If a direct implementation of the algorithm takes a few million qubits, it may well be that both better algorithms, better error correcting codes, and optimizing their combination to the hardware platform will shave a few orders of magnitude. So in 5 years to a decade this could get uncomfortably close to the danger zone. Main issue is that in cryptography (1) some secrets encrypted now will be secret in a decade (2) changing standard is very hard and takes a long time