I believe
(though have not verified it myself since I utterly despise the
machines) that Sinclair used this technique in the early Spectrums; they,
apparently, simply disabled the faulty half of the DRAMs using jumpers on
the board.
I beleive (although, like you I've not checked because, also like you, I
have as little to do with Sinclair machines as possible) that early 48K
SPectrums had 8 16K chips and 8 32K chips. The latter were half-good 64K
chips, there was a link (or maybe a resistor) on the PCB to select which
half would be used. Needless to say all 8 chips had to ahve the same half
good.
That's right, they did. I have the workshop manual somewhere that details the
jumper settings for all the variants.
IIRC surplus 4164s worked in any machine...
Gordon