I'm not sure I believe your benchmarks, honestly. They are way slower than the C++ version. But I don't really have a horse in the race as simdjson-go is its own thing. Still, I'm mildly surprised that a byte-by-byte parser is the same speed as a SIMD approach, and usually when someone tells me this, the answer is usually that they buggered up the benchmarking. I would suggest comparing your numbers to the simdjson paper or whatever brag numbers simdjson-go has as a reality check; one of us is clearly wrong.
The use of parallelism in simdjson is the use of SIMD instructions, not multiple cores. So the answer is "no" to the second question.
The use of parallelism in simdjson is the use of SIMD instructions, not multiple cores. So the answer is "no" to the second question.