Game oriented machines of '80

Amstrad CPC has some really colorful games, but at the cost of lower resolution (160x200 16 colors) than MSX1, and to me, all the CPC games are slower. The Z-80 runs at 4 Mhz, the locomotive basic is better than msx basic but that´s all. MSX 2 is way better than any other 8-bit computer

lol screen3 is lower but color is very good

the situation is not that worse as you describe. on MSX1 accesses are tipically contiuguos in pratice you set the vram ptr then read a block or write a block of contiguos data. This way the initial overhead is lower. In the worst case scenario effectively you need to compare about 80 cycles to do the same you do in direct accesses with a LD A, (HL) that takes 8 cycles on msx.
On other situations this kind of access can be faster. Take for example a transfer operation where you need to track both source and destination without using LDIR, with the OTIR solution you can free a pair of registers for others uses.

For the other problem ( you have to write color on second run ) this is also true on ZX, the difference is that you need to write 8 times colour data than on ZX. And is this that really slow down the game. If you check the batman game the zx convertion is fast enough and comparable to speccy because there is no color overhead.

writing 8 consecutive color bytes to emulate the speccy can be achieved in vblank area in this way and is the faster way.

call setvramptr
ld a, colourcode
out (0x98),a
out (0x98),a
out (0x98),a
out (0x98),a
out (0x98),a
out (0x98),a
out (0x98),a
out (0x98),a