It might have to do with the raster position at which your bitplane pointer writes execute.
If I'm reading the copper list correctly, you set up the display fetch & window first, then do a copper wait, then bplcon values and write colours and only then write the bitplane pointers. These may be slightly too late as the writing of AGA colours takes a while (you used 64 colours IIRC, so that'd be 64*2 moves = over 2 scanlines).
It may simply be a case of moving the bitplane pointer writes to before the colours.
Edit: same idea as Ross, different way of putting it
Edit 2: a way to get a grip on these things, I find, is to use the WinUAE visual DMA debugger. This shows the position of the copper activity vs the bitplane fetching so you can see if you're still writing during display (and where on the scanlines)