Yeah, primary or aux sensing is popular with low cost, small supplies -- the regulation is ass, due to leakage inductance (which is on the poor side as well, due to the simple transformer windups they use, presumably for cost), but it's good enough for general purposes, phone charging, etc.
Electrolytics really are the dominant failure mechanism; beyond that, probably transients, anything to do with marginal design ratings, insulation quality, etc. All of which can be designed for adequate margins, protection, filtering, etc., it just takes a little more work to do.
For their part, semiconductors really are remarkable. A rock that's essentially unchanging, yet capable of highly dynamic and intricate behavior. They can last pretty much forever, as long as they're designed for it. (And for that, electromigration is probably the biggest barrier to long life of most commercial grade products? Older designs probably win here, as the wider conductors have less current density and thus susceptibility to electromigration.)
There's plenty of examples of old consumer hardware that's survived despite its cheap design; sometimes you just get lucky. The challenge from a design standpoint is getting a 90% yield after say 50 years, rather than 0.1%.

Transients bear some reflection. Keep in mind, we normally test (well, if/when we do..) to a standard level, and that's that. It's designed as an upper limit level, so that anything seen in practice likely won't exceed that, but there may still be cumulative damage, and there's still the outside chance that a worse event comes along, just too infrequently to matter with respect to the test. And that's good enough for commercial lifetimes, maybe a once-in-10-years transient exceeds that level and blows it up, and that's fine for commercial equipment, but if you're targeting longer life, you'd want to test to higher -- and use protective devices that don't wear (those stonking great mains TVSs), or at least much larger devices that wear less.
Tim