My understanding was that if you're doing random access - ZRAM has near-zero overhead. While data is being fetched from RAM, you have enough cycles to decompress blocks.

Would love to be corrected if I'm wrong