>what if we turned this all on its head and intentionally designed a hash function that is "ASIC optimized" so that the perfect ASIC miner design could be found quickly

Since hash functions are applied to variable-sized parameters, optimizations always exist. You can always compute the hash by processing less of the input, or by processing it in parallel. Just because it's not known how to do this doesn't mean it's impossible. It's impossible to know when an implementation of an algorithm (whether in software or hardware) is optimal.