The trouble is real world applications (like combining Whisper + piper + HA, etc.) differs from specific benchmarks (many of which target different parts of the Linux stack that may be more optimized for arm or x86 specific features...).
And that doesn't account for individual N100 device variation—some have different power and thermal capacity (e.g. the Radxa X4 is limited compared to a larger mini PC).
And that doesn't account for individual N100 device variation—some have different power and thermal capacity (e.g. the Radxa X4 is limited compared to a larger mini PC).