That’s right, this is most likely a firmware bug or a flaw in the Rigol DHO800-900 series!
I carried out the same test measurements on the Rigol DS1054Z and the maximum phase deviation, regardless of whether 2, 3 or 4 channels were included, was about 1.2 degrees on the test signal - Sine 70 MHz.
Here is the phase shift of the signals when two channels are turned on:
And here is the phase shift of the signals when all four channels are turned on: