On GPS-locked E7495A and ERA tuned to 10MHz TCXO seems to be spot on, surprisingly OCXO appears to be 14Hz low. The narrowest RBW is 30Hz, so it's not conclusive, but I did not expect such a big difference, especially for OCXO. Unfortunately I don't have GPSDO to lock ERA, so I can't tell how much of this difference comes from the reference XO, and how much is due to poor resolution of the instrument.
Update:
The same holds for ERA set to 2.5GHz. TCXO is much closer to the set frequency than OCXO. Frequency locked to the OCXO is ~7kHz low, and TCXO is ~1.3kHz over. If it wasn't for frequency drift while OCXO starts up, I would tell that the labels got mixed up. Also, the frequency in steady state seems to be constantly drifting by single Hz, but again, I can't tell how much of it comes from the SA itself.
Last, but not least, refreshing the diagnostics page during OCXO startup shows all three PLLs as locked through all the time as the crystal heats up. I did not expect the Xtal PLL to respond that quickly to sudden changes of ref frequency.