It's late here in the UK and my brain might not be working correctly so I may be about to make a fool of myself but...
Two frequency counters with an internal 10MHz reference that's brought out to a BNC on the rear panel (Racal-Dana 1998 for instance).
Connect them together so counter A's reference is displayed by counter B.
Add another cable so counter B's reference is displayed by counter A.
Adjust A and B reference oscillators until both displayed frequencies coincide.
Result, both reference oscillators are the same frequency and should be extremely close to 10MHz?
Or am I missing something subtle?
Or worse, something blindingly obvious?
FWIW, I've just bought a pair of 'faulty' 1998s, both now repaired but in need of calibration. I'll be buying a Rubidium oscillator as soon as funds allow so should be able to calibrate them against that next month but as a thought exercise the above occurred to me.