Hello,
Sure it would be helpful to get comparison with same algo...
stak mines with 3 threads with 2M cache each, for your total of 6M cache, which sounds good.
What happens if you mine with the same configuration with JCE ?
multi_hash:1 on three CPUs: 0, 1 and 2?
I'd expect you get at least on par results, not -40%.
Your current config using only 2 cores and 8M cache (2M overflood) is expected to be bad.