The resistors are Chinese metal film from the same lot, binned with a 5 digits DMM. The 10k in the base were +/-1 ohm (10.013, 10.014) and about the same tolerance for the 1k resistors in collector.
As a double check, when the transistors are swapped the DC offset changes its sign, probably less than 1-2mV is caused by resistors mismatch, while the best matched BJT pair found so far produces about 9mV of DC offset.
The TC of the resistors seems good enough. When touched with the finger (through a plastic bag) the offset stays about the same.
The transistors, however, go crazy with the temperature difference. I knew Is changes with temperature, but I never measured how much would that mean in practice. By simply touching the capsule through a ziplock bag, the offset can change by many tens of mV.
The pair of BF458 (TO-126) I am measuring now started with 8mV DC offset, slowly raised for about 5 minutes, and reached thermal equilibrium at about 40mV. But if I touch one transistor the DC offset slowly grows from +40mV to 100...150mV. That's more than 300% offset increase!
If I touch the other transistor, the DC offset changes with about the same amount towards the other direction, to negative values.
Now it makes sense why precision opamps are designed to have a thermal axis of symmetry, and with the critical components carefully arranged with respect to the thermal axis.