Fair point, although with a 400+ year half life in the americium source in the detector, I am skeptical that a new smoke detector would be any more reliable than a very old one.
I would think testing them regularly - especially with simulated smoke as done in professional situations, or in my case via bad cooking, is probably more effective than regular replacement on a schedule to ensure they are always working.
If dealing with something that follows a Poisson failure probability distribution with a fixed percentage probability of failure per year (as is the case with most electrical components), regular replacement only makes the system more reliable if you are unable to test it, otherwise it makes no difference.
With a few rare exceptions, is largely a myth that replacing machines or technology at regular intervals increases reliability- people incorrectly assume this to be true, based on observing that most failures happen to things that are old, but this is merely because they spend more time being old, not because the rate of failure per time increases with age (it almost never does). Testing and redundancy are more effective and cheaper.
Now, everything I am saying would be wrong if smoke detectors indeed have components besides the alpha source whose failure rates are known to increase with age, and actually age out within a decade or so. Like you mentioned, this can be the case with electrolytic capacitors as well as non solid state relays. However, I wouldn't be surprised if the lifespan of capacitors at the low temp and low voltages in a smoke detector wasn't 50+ years.