I wrote a code that uses BMXL to copy from linear memory to coordinates on the v9990. This is what I get on my SMX with the Power Graph:
This is P1 mode. A blocks are layer A blocks. B blocks are layer B blocks.
But this is what I get on OpenMSX (exactly same code):
As you can see, the orange blocks are not visible on the real hardware. They are there just to fill empty space so we can see what BMXL is actually doing on OpenMSX. So it seems that if I copy memory on the real hardware, BMXL uses half the value in NX compared to the same command on OpenMSX. Here is the program if you are feeling adventurous.
I will provide source code after I clean it up a bit.
