This ignores potential current unpublished advances by entities like NSA and potential unforeseen future algorithmic and computing power advances. Logically speaking, larger keys could help and are unlikely to hurt with both. Even if larger key still ends up breakable, adversaries may still go for low hanging fruit.
Other than that, it depends on secrecy timeline and cost/performance sensitivity. An average credit card transaction is unlikely to be targeted by NSA or archived in hopes of cracking it 30 years later, and on the other hand volume is very high and latency is important. So use whatever is thought to not be breakable now and upgrade keys if and when technology progresses. On the other hand, list of American spies in Russia would not take more than a few minutes to decrypt even with enormous key sizes and on the other hand disclosure could cause real damage even decades later. Might as well overshoot even if there is no known reason as of yet.
Other than that, it depends on secrecy timeline and cost/performance sensitivity. An average credit card transaction is unlikely to be targeted by NSA or archived in hopes of cracking it 30 years later, and on the other hand volume is very high and latency is important. So use whatever is thought to not be breakable now and upgrade keys if and when technology progresses. On the other hand, list of American spies in Russia would not take more than a few minutes to decrypt even with enormous key sizes and on the other hand disclosure could cause real damage even decades later. Might as well overshoot even if there is no known reason as of yet.