You could make mining more memory bound rather than compute bound, somewhat reducing electricity costs and requiring bigger upfront investments (in DRAM chips). This also vastly reduces the performance gap between commodity and custom mining hardware.
Why would that reduce costs? Miners would have to compete for memory & CPU then.
Even altcoins that have tried this approach have seen custom hardware appearing if the coin grows popular enough.
The approach just seems to be a desperate hope that home users will be able to mine profitably, but ultimately it still fails because those with more resources will win out.
It shifts, rather than reduces, costs.
No altcoin has used a PoW where memory latency dominates computation.
Home users will never be able to mine profitably (except by claiming their electricity is "free"), but the hope is they can do so within an order of magnitude loss of efficiency.
You might be able to make mining more efficient, or based around some other finite resource besides computation power (like storage), but even that's a long-shot.