Yesterday, I took a new mixer (not the old TAMA one) and monitor its IF channel with two identical frequency RF signals as RF/LO.
The result is attached. Comparing this monitoring with elog2616, we can see much smaller drift.