Yeah, there's a lot of weird stuff in the Lisa.
The most blatant are flops that can be enabled or disabled on reads. These are used to turn on the MMU contexts and memory parity checking tests, for example. That's right, a read is actually equivalent to a write, because it only cares about something touching an address, and not whether the write line is turned on.
Another one is that one of the lines for the parallel port is actually attached to the keyboard VIA instead of the parallel port - they could have fixed it before they released hardware, but chose not to. This isn't the case for the expansion slot dual parallel card.
Another is that they claim the VIA 6522s can push 625KBps - they actually can't - they lied. What they did is divide 5MHz by 8 clock cycles to get 625000. There's no DMA for the VIAs. So basically the ProFile driver would have to read a byte from the port, turn around and write to memory and increment the address register, and repeat 512x. This also doesn't include the fact the the 68000 has to fetch instructions from memory for each opcode, and there's effectively no cache.
So yeah, whole lot of slowness everywhere.
Funnily enough there's two coprocessors built right in, a 6504, and a COP421. There's also the possibility of adding an AMD FPU or alternatively a WD2001 DES encryption co-processor on the 2/10 I/O board - however, it appears that there's not much significant increase in speed with the FPU; I'm unsure whether the FPU had support in the OS for it or not. There is a SANE library disassembly somewhere, I don't recall anything there looking for the FPU, which is a shame. The FPU communicates with the 68000 through the VIA, (I'm assuming) making it slower instead of being directly on the bus.
A slightly better design would have been to have the 6504 talk to the keyboard and mouse as well as the floppy, but not sure it would have been able to provide a real time clock, or if it had enough free I/O port lines for that. It's also likely the COP421 is lower power than the 6504 so maybe it would have needed more amps on the +5V standby.
(Pointing out here that a microcontroller + mux + extra VIA were needed, when with a bit more optimizing the 6504 could have done some of that heavy lifting and most of the time the floppy drive is idle.) Certainly they could have ported the floppy code to the 68000 and not needed the 6504 at all - the Mac directly accessed the IWM from the 68000, etc.
So, here we see a lot of wasted hardware as well as some really weird optimizations for the flops that control the MMU. (Like anything else this just points out that it was designed by multiple designers, some more advanced than others, some who optimized for fewer chips, others for a more mini-computer design.) The same kind of stuff can be found in the Boot ROM source code, and certainly within LOS itself.
The ProFile hard drive protocol is pretty annoying too, and still gives me trouble today. I really hate working on the Profile emulation code for LisaEm, but yeah, that's what I'm doing right now. Widget has an improved protocol, but introduced even more weird things, but at least it has a somewhat more simple protocol.
For example, for ProFiles the tags go first, but for Widgets, the tags go to the end. To do this and still allow for LOS to boot, they did a hack by adding a JMP (might have been a BRA, I forget) in the boot code on the Widgets, followed by AA,AA which allows the Boot ROM to still boot from this new scheme, and then the tags go at the end.
The actual drives don't care about where the tags go, but they seem to have done that to get a slight increase in performance for when they don't care about the tags, they can be discarded.
But, LOS does care, only MacWorks doesn't, so, I still don't get why. Even worse, the boot loader is different between ProFile and Widget and you can't image one over the other to use as recovery. (Not that you could back in the day as no tools existed, though now they do.)
Still even worse, if you install LOS on a Profile off the motherboard parallel port, and then move it to one of the two ports on a dual parallel slot card, it won't boot - somewhere in the boot process it loses track of what device it started from and throws something like 10707. There's a binary file that's used to store the configuration and that hard codes the device, so after the kernel loads it reads that file and then that file says, go read the rest of the OS from the parallel port, and fails.
The funny thing is that there's only a single driver for both ProFiles and Widgets, and if you have a Widget, it will move the tags to the end of the sectors, while if you have a ProFile, it will keep the tags at the front of the sector. Why?
The classic screw up was making QuickDraw non-rentrant, so the multitasking had to be cooperative. This might have been fixable with a single semaphore or lock. D'oh.
But despite the hardware and software weirdness, LOS is a hell of a lot better and more mature than the original Mac System 1.0, which was pretty difficult to use until ~512KE with an HD20 (or at least a 2nd floppy drive.)
It's too bad they threw LOS away instead of slowly porting parts of it to the Mac as it went to things like the SE/30.
Examples: they could have used the Lisa's file system which had folders, but went with a silly one: MFS, which was later replaced by HFS. Lack of an MMU until the 030s, which meant, lack of virtual memory, lack of pre-emptive multi tasking all the way to MacOS 9, not remembering what windows were open when you shut down, hard switch power off, and so on.
MacOS certainly got to be very good by the time it got to System 6.x, but even then, LOS still had a lot going for it.
The one really neat thing MacOS (well System) had was the Resource manager which was really a brilliant invention and ResEdit made it awesome. It would have been even better if they had turned it into an actual database (there's even a TechNote warning people to not use it as such - sad.)
Also MPW is a much nicer IDE than LPW. LPW doesn't really use the GUI for anything, though the text editor is a GUI tool. Speaking of LPW, it has a scripting language that's a hack. It looks like originally they just accepted keystrokes from a file instead of the keyboard as a scripting mechanism, but later extended it to a scripting language but did it very poorly, so it's half of a keyboard redirect and half of a scripting language.
The Mac XL certainly was my first Mac, I used it with MacWorks exclusively for years, never knew about LOS until much later, and then learned Unix via Xenix on the Lisa.