I'm not running on a Pi4, but I am running on an N150 which is in roughly the same performance league, and CLIP searching is instant in my ~10K photo set. The expense is at classification time, not search time. Classification was a few hours on that, so, not convenient if you're staring at it and wanting to play with it instantly, but it's not like it took months or something either. Of course if you've got 100,000 photos it may be some days for an RPi, but it's still just something you can let it crank away at, it's not like you have to stare at it while it happens.