My 2 cents...
For me there are 2 interesting ways to think about this topic:
1. a pure hobby project to create a computer that *would* have been the MSX3, if MSX wouldn't have flopped and everything would have been done properly and ready on time etc. What would that MSX3 look like? So, a 1992 (or so) computer that would have been commercially launched as MSX3. I think it would have had V9978. But that doesn't exist, so best thing that comes close to that is a V9958 superimposed on top of a V9990, I think. It would have MSX-MUSIC, probably. What else? OPL2? Probably! It would also have a faster CPU, but also be totally backwards compatible to MSX2 and MSX1. So, what do you think, how would this MSX3 have been?
2.a modern computer with the "MSX spirit" (whatever that is), which is not per say backwards compatible to any MSX. This could very well be a modern OLPC laptop, as was actually already hinted on by Nishi in 2001. The spirit is then something like: quickly on, easily programmable, inspires users to creation of software, graphics and music.
I would be interested in that MSX3 as I sketched above at point 1 
A fair upgrade done by the MSX hobbyists, because the commercial world failed it
This fragmentation can also occur if I finally manage to convince a hardware genious to develope new gfx9000, as happened with the opl4 being manufactured today.
You mean you conviced someone in Spain to create a new Moonsound Clone? Can you tell us about who made this and what the specs are (SRAM amount, special features etc)? And if we can still order it? I think it would help the international msx scene since there may still be people looking for a new one.
I already have an opl4 cart but I'm always curious about new msx hardware 
That I understand. That would be cool. Except that GFX9000, although it existed it's rare and could increase fragmentation. Maybe cool feature, but don't write too much GFX9000-ONLY software please
Its not that rare, as Grauw says tecnobytes have been making them on and off (as in small batches once every few months) for the last 5 years or so.
Still it doesnt increase fragmentation much because even up until now not much (close to non at all?) new stuff was released for it, and frankly, I dont see it ever becoming a "big" thing unfortunately. I think a recent increase in interest in the gfx9k is because of symbOS. Gameswise I only know that assembler is working on some gfx9k things. Hope they get released at some point :D (I personally would like to make something for it... but first things first:D )
There’s a Graphics 9000 Basic which was supplied by Sunrise together with the Gfx9000 and should still be available for download on their website.
Edit: ok, you were talking about a “different” assembler haha .
I would like to do something with the Graphics9000 someday but at the moment I have no specific plans. I’m trying to think of something that will provide a nice benefit for Gfx9k owners without excluding other MSX users. A multi-engine game perhaps. Or a Gfx9000 version of some MSX2 UI so that MSX1 users can also get pretty graphics.
On-topic, all I’m really looking for in a new MSX is a turboR reproduction with V9990 built-in. TurboR is a very capable and fast machine with smart architecture, but its cost is a big barrier for ownership. A “new turboR” would increase the turboR user base by a lot, would have existing software support and also developer support out of the box.
As for why to add the V9990, it is the only hardware expansion that doesn’t really work well as a cartridge (multiple video outputs is a hassle), and the V9958 has always been the turboR’s achilles heel, so I think it makes sense to include it together with the V9958, despite its low software support at the moment. At least it provides a meaningful “upgrade” to turn the turboR into a true MSX3, along the path (presumably) originally intended by ASCII, without actually introducing new hardware fragmentation to the MSX market that would discourage us developers from using it.
I strongly believe that we should forget V9990 and R800, to determine the spec.
Both legendary devices cause descent of possibility of realize.
Functionality, performance, availability, and cost, V9990 is not advantageous in all of them.
Also as R800, it's no longer obtained easily.
We should not be swayed too much thought of backward-compatibility for past rare product.
Yes ofcourse, if your goal is a machine that moves BEYOND the turbo-r+v9990+opl4 concept then, imho you don't really need to keep backwardscompatibility with v9990, it just makes the pcb bigger and more complex i guess... People who want to use v9990 software on your machine can just get a gfx9000 or clone 