Hello again,
I have finally found the issue with the SELFOCAL command. The time between the last falling edge of SCLK and the slave deselection (/CS being pulled high) was not long enough. The datasheet specifies it has to be at least 7*tosc, which equals 1,71 us. I did not have a sofwater wait, so in most cases that criteria was not satisfied.
Thank you very much for your support. Kind regards,
Juan