Constantly growing caches just work very well, and we've grown to treat the DRAM level of the memory hierarchy as just another storage device, faster than your SSD but slower than your L4 cache. The industry has just emphasized capacity/$ and traditional, commodity DRAM protocols. Consider for example the potential, and near-absence[1], of CPU + DRAM MCMs.
[1] The latest rebranding of this obvious idea is "HBM" in GPUs. Though these don't really take advantage of the idea. But still CPUs have not been interested in even reaching this little distance.
If Smalltalk VMs had the same resources thrown at them as other VMs/compilers/etc. they could be more competitive speed-wise. Noting, of course, that Smalltalk systems can be plenty fast-enough.
Smalltalk-as-glue is also perfectly reasonable, interfacing with multiple back-end systems, controlling business logic, etc.
It's true though that a parallel object language like Concurrent Smalltalk isn't the historical Smalltalk we usually mean.
There may be some changes to language semantics needed of course, but doesn't preclude Smalltalky languages.
Another property of the current software/hardware world is that the vast majority of programs don't care about performance at this level. There is a vast size difference between the oceans of production code in the world, and the amount of "hot" code that people pay money to make faster.
(And the levels of "make faster" are usually more low hanging fruit than switch of programming language)
> but it was designed in an era where hardware had very different performance characteristics from today's hardware.
Exactly, the era of 256k memory computers are over. Programmer time and correctness is more valuable than making sure everything fits in L1 cache.
Given how much time your average huge application (Word, Photoshop…) takes to even start up, I think this is still an important point.
Programmer time is important, but so is user time. And in many cases, there are many more users than programmers.
I believe the pendulum is swinging back from dynamic languages to compiled languages with stronger static guarantees. Languages that are parsimonious with memory and give you control over the layout when you need it.
Don't get me wrong. Smalltalk is a hugely influential language and we're better for having it, but it was designed in an era where hardware had very different performance characteristics from today's hardware.