You technically can make an ASIC for anything, but... in some cases it's just plain stupidity. For most algos, ASIC-capability was actually a design criterium, so I won't blame them for being implemented. For some, the memory constraints alone make it a less interesting proposal (scrypt for example, with scrypt-n being a further development), but it's still possible. However, there are quite some algos that for now run mostly on CPUs, even with programmable GPUs all over the place, and enough greedy developers to get it working. Why? Because you'd have to design the whole thing from scratch, using functions that are very very slow on a GPU (random memory access, pointer chasing) and probably by the time you finally made the super-ASIC for it... the algo is out of favour.
Personally I will not focus on scrypt ASICs, I'll just move on to the next type of coin with my GPUs. The ones deciding the value of coins (those who spend money on them) will follow suit, probably. Pay2win has never been the idea of cryptocoins, even though that became reality the past months... The problem with pay2win is that someone else will throw more money at it than you can, and you're out of the game.