Perhaps I found a major bug trace. Using kickstart 2.0 raw cli,
dragging the screen down will shift or "destroy" bitplane below
line 200 (NTSC range).
In Kickstart 1.x it is working correct, most likely 2.0 use a
different copper list.

I already change the copper and got a bit better result, also in some
demos it shows little improvement.
Still this instruction $FFDF,$FFFE could not work right or will be
ignored?
Would be nice if somebody could crosscheck this.

digest of copper.v in this matter (very last part):

//generate request dma (reqdma) signal
//for a dma to be requested first of all the cycle must be right
(horbeam[1:0])
//second, selins or selreg must be true (state machine request bus
operation)
//the last cycle in a line is not usable by the copper
always @(selins or selreg or horbeam[1:0])
if( (horbeam[1:0]==2'b01) && (selins || selreg) )//request dma cycle
//(horbeam[8:2]!=7'b1110001) &&
reqdma=1;
else
reqdma=0;