Oh, and I forgot to mention the important part. I've used those 2-bank 128MB SIMMs in the Q605 and they work fine.
I posted a bunch of test results either here or over on Applefritter a year or two ago. My results tracked Lawson's pretty closely, which makes sense. I did not have an LCIII to test, however.
The QxxxAV and the PM7100 and 8100 only seem to support 11 X 11 memory parts. So the largest possible bank is a 11 + 11 = 22 bits => 4M addresses X 4 bytes wide = 16MB. So a two bank SIMM with 16 MB/bank gives us a maximum supported 32 MB SIMM.
The PM6100 does support 128 and 64 MB SIMMs even though it uses exactly the same memory controller chip as the PM7100 and 8100. This puzzled me until I used an ohmmeter to trace the connections. The PM7100 and 8100 run their address lines and their RAS and CAS lines through a largish PLD before it goes to the SIMM sockets. I don't know what happens there, but part of it is that support for 12 X 12 memory chips is lost. I suspect that chip steals the upper address bits and uses them to control access to the different SIMM slots.
Anyway, that's why the 6100 behaves differently than the 7100 and 8100. However, the report on Kann's website of a pair of 256MB SIMMs in the 6100 is almost certainly false. There are four CAS and RAS lines available in the SIMM socket and the 6100 ties them together in pairs. So there are only two available to distinguish between the two banks in the 6100 SIMM socket, and the lines are used 1 hot, not binary encoded, so 2 lines give 2 choices. There does not appear to be any way to build a 4 bank SIMM which would be supported in the 6100 and given that 64MB is the largest possible bank capacity, a 256 MB SIMM would require four banks.
Also note that some 128MB SIMMs, (the HP part I quoted above, e.g.) tie their RAS and CAS lines together in the opposite pattern to the PM6100 and so those SIMMs won't work at all in the 6100, although it might be possible to change the RAS tieing by moving resistors on the SIMM.
The 610, 650 and 800 see half the capacity and so apparently do support 11 X 12 or 12 X 11 addressing, because they treat 12 X 12 memory chips as 12 X 11 parts. So, in theory, one could install 128 MB SIMMs in those machines and get 64MB of capacity recognized. I think. Surely I tried that...I need to find my old posting on this topic.