Why? Because 6502 is a poor man RISC processor with Von Neumann bus interface.
......
As discussed in many threads, V9938 was a silicon prototype and obsolete device when came out.
1) Maxis do you think they have in mind the v9958 when they released v9938?
should be interesting to find some evidence on v9938 that prove that v9938 is something like a BETA v9958.
2)please do not call 6502 a RISC processor. RISC mean reduced instruction set computing. Reduced not in the number of available opcodes, rather in complexity of the opcode itself. And i cannot consider a pre/post indexed addressing operation a kind of simple opcode. from this perspective it's more RISC the z80 that 6502.
the maximum we had was a ld reg,(ix+2) ;-)
6502 is a low cost cpu, stackptr of 8 bit for example.
only 3 registers A, X, Y, zero page memory oriented cpu.
1. Yes, by looking at the VRAM timings (time slot access reservation) I can only say that either Yamaha wanted to reduce the number of the silicon pre-prototypes by integrating and not using some features in V9938 and "flight proove" the next generation, or the V9938 design project was running very late and the only option of making it on-time was to cut some features at the last moment. Hence - no time to cost/performance-optimize the design, which is a common design phase, which comes after the silicon prototype validation.
2. When we define the RISC to the students we give the clear case example, which is Standford MIPS processor.
ARM which stands for Acorn RISC Machines, has also post-indexed and pre-indexed modes close to 6502. Also ARM including ARM7TDMI doesn't exploit the Harward archicture either. And ARM is still the RISC processor. Why? Because the main criteria of being a RISC or CISC is having or not the PLA instead of the microinstruction sequencer. So, the definition of the RISC in the industry is not such a clear case and actually more related to its implementation or microarchitecture.
The presence of PLA revolutionized the 6502 based computers and allowed to interleave the CPU bus access with the graphics/DMA bus access without slowing down the latter. Hence, 6502 is a poor man RISC.
Z80 couldn't provide the same. Hence - separate video RAM was the performance improvement decision. The well known attempt to combine the fast CPU access and memory mapped video buffer as we know was made by Sinclair Research in their Spectrum 48K. Spectrum 48K has 2 memories: fast RAM for the CPU only and slow RAM for the shared VIDEO DMA + CPU access. And the bus management is solved by ..... 8 resistors and one TTL chip. Brilliant, but this is an analog rather than digital solution.
For the others -> the separate VRAM became the main choice. Also we remember, that MSX1 was a very successful computer having no custom chips specifically made for MSX. So, the choices were limited.
Sorry for the offtop 
