From the release notes of the 00.01.00.00.25 firmware, they "fixed bandwidth limit issue and optimized filter performance".
I performed the bandwidth tests again and could verify that the bandwidth at 1Gsps is no longer significantly below the specified 400MHz. However, the bandwidth at 2Gsps is now lower than the specified 800MHz. I have filed a bug report to Rigol.
The good news includes that the amplitude responses under different sampling rate conditions are now reasonably matched within the respective pass bands. Therefore, the captured waveforms are now less impacted by switching analog channels on and off, as observed in the earlier firmware versions.
Also, the ~1.2GHz filter at 4Gsps now has a slower roll-off, and there is less ringing in the step response.