On all CRTCs, there is a 1µs delay in display between when the CRTC provides a video pointer, and when the Gate Array displays the corresponding 16-bit character.
But on CRTCs 0/1/2, there is no delay for HSYNC. On CRTCs 3/4, the Amstrad engineers fixed the issue by adding a 1µs delay for the HSYNC signal to match the video signal.
So now we have a bigger issue: on CRTCs 3/4, HSYNC occurs 1µs later than on CRTCs 0/1/2. Interrupts being dependent on HSYNC, this is a serious compatibility issue for time-sensitive code. It also explains why the CTM monitor has to be calibrated differently on CRTCs 3/4. Fortunately, the issue is easy to fix, by adjusting the HSYNC width with CRTC register 3.