Per watt, Raspberry 4 is probably peak CPU of our civilization. But 64-bit is only viable if you buy the twice as expensive 8GB model and have use for the extra bandwidth. I'm building hybrid clusters of 4 (2GB) & 2 (the older version) on 32-bit so I can run my PaaS at record flops/W/$: https://github.com/tinspin/rupy
Benchmarks don't entirely agree: https://www.cnx-software.com/2016/03/01/64-bit-arm-aarch64-i...
If the difference between AArch64 and AArch32 were simply larger pointers, I would probably agree, but AArch64 includes a number of instructions that can do more work, so 64-bit programs can actually run faster than their 32-bit equivalents, even if you're not using much RAM.