Qwen3.6-27B benchmark: heavier quants boost speculative decoding

Benchmarking speculative decoding on Qwen3.6-27B, the user found heavier quants gain more speed: all 10 speculative configs ranked Q8 > Q6 > Q4 by multiplier. Token acceptance stayed quant-independent at matched depth, while the base step slowed as quantization weight increased.