That's what I meant: for performance considerations, L1 latency must be limited to single digit cycles on a general purpose CPU, hence L1 size is capped.