Tested in openMSX, not yet on real hardware. The screenshots are of the patched ROM running in openMSX (an MSX1 and a Philips NMS 8250), and the numbers come from a probe that counts, frame by frame, the sprites the VDP does not draw. No ROM is distributed: only the IPS. It comes from the commented disassembly, where every address can be looked up.
Every frame 0x5E1B sorts the seven groups by depth, 0x5C0B builds their attributes in that order, 0x5BAB copies them reversed as well and the interrupt (0x5B77) shows one table or the other: one frame the near players are in front, the next the far ones. On an MSX2 the patch always leaves the sorted one: the nearest in front, and no change from frame to frame.
Register 5 points at a 1 KB block: the sprite colour table, 16 bytes per sprite, takes the first half, and the attributes sit 0x200 further on. Put them the other way round and the screen has not a single sprite, which is the first thing that came out.
Below the 16 KB of the MSX1 the game uses nothing from 0x3C00 to 0x3FFF: colours at 0x3C00, attributes at 0x3E00 and the MSX1's second table at 0x3F00. Not one pattern moves, and the same file works on both MSXs.
The filler at 0xBFD4 and three routines nobody calls (0x6EA0, 0x6F30, 0x7AAE), found when the disassembly was traced. And the sample figure of the team menus keeps its colours: mode 2 does not read the attribute's colour, so the patch copies them into the colour table.
openMSX screenshots of the ROM each caption names; the commands that take them are in In the emulator.
0x6F30.