Actually, you do not really need the speech center of the brain if you can display words on a screen. So you can leave at least that behind. And computers get exponentially more powerful, so that's not a huge problem.
Computers getting exponentially more powerful isn't the issue. The issue is that there's more than 1 bit of data contained in an atom, and you're going to need more than 1 atom per bit of data you want to represent. So you can't
properly model a chunk of matter in less matter than the chunk holds and/or in less time. e.g. to model 1 second of existence, completely, for 1 gram of matter will require much more than 1 gram of computer, and/or much more than 1 second. No matter how fast computers get. This is a hard bottleneck for simulation.
e.g. if you can
jettison large amounts of the actual information in the brain and still have it completely functional then you have a shot. If however, the brain turns out to use a whole pile of quantum effects, to the point that a lot of the matter actually matters, then you won't be able to "properly" simulate it in less matter than just having a brain to start with. It all depends on how "compressible" consciousness is.
e.g. you mention saving space by jettisoning the speech center of the brain, but that's just wiping out a large chunk of actual brain functionality. Is the experience of being the brain-jar really going to be the same if we remove the speech centers completely? That would seem to make the simulated brain-jar already an imperfect simulation, if we resort to things like that. There's a
ton of feedback between other brain areas and the speech center, it's not just an "output module" that can be discarded.
So it comes down to it that you
could make computers powerful enough to simulate a brain, but not necessarily
cheaper in resources and energy than just having a brain to start with. Which goes back to your original point:
why have brain-jars when you could "just" upload them? Because the brain-jar
costs less in terms of resources and energy than running a complete simulation of the brain-jar.