This is back when memory was actually expensive, so yeah that is pretty big.
At the time, there was much frustration that Apple
Cambridge spent so much time on their new development
environment, and not as much as we wanted on making
improvements to the Dylan runtime (for example, we hired
Ken because we concluded that Cambridge would not get
around to threads in time to matter to us). That said,
I have nothing but respect for what Apple Cambridge
accomplished.
Given that, it is pretty likely that they lacked a tree-shaker to reduce executable sizes (https://groups.google.com/forum/#!topic/comp.lang.lisp/pspFr...). It is also pretty unlikely that they'd spent enough time on reducing the memory usage and so on. (It is worth noting that even today, Open Dylan hasn't got a tree shaker.)
Bringing up a whole new ecosystem from scratch is hard work!
I think the switch was due to resource constraints (512K of RAM was a big deal in a handheld product) as well as a sense that they needed to get real and ship something. The impetus may have been a small group of people (e.g., Capps and Walter and some other heavies) having a Serious Talk while at a trade show in June of 1992, but obviously management must have been involved, too. Anyway, I joined the group the week a bunch of people were out of town, and when they came back everything was different.
There were a number of sad-faced LISP hackers roaming the halls, clutching copies of the C++ Annotated Reference Manual. The ARM (the book, not the CPU) must have been hell for them.
From that point on the Dylan stuff was pretty much sidelined. The "Senior" product (a tablet) was kept alive for quite some time, but the group's focus was almost 100% on Junior. There were occasional releases of Dylan (and I got a copy of the book), but little true interest.
Paul Mercer went an extra step further in reducing risk and complexity; he proposed ditching the microkernel in favor of a relatively simple a 68K-based handheld running an event loop and some apps, and he even got a lot of it working in a demo on some prototype hardware he whipped up. This was a little too radical, since there would have been no handwriting recognition, and he wound up leaving (and later founded Pixo, which was contracted to do the first UI for the iPod and was eventually bought by Apple). I'll note that what Paul proposed is essentially what the Palm Pilot did, with a limited character-by-character recognition interface that actually worked pretty well.
In any case, as Alan Kay says:
"if you want to see the future, you need to be willing to build the hardware from 10 - 15 years in the future."
Xerox PARC systems would never had been built if they would just use PDP like systems.