For a little bit of insight into how the professional repairers would do it...
I'm not a professional repairer as such, but was doing it for money at one point. But anyway, looking at a BZ board with lots of faults at the moment. As Mike says, first thing I do is to stick it on the Fluke (9100FT in my case) and test the RAMs. In this case the RAMs are all 2114's, but same principle.
First up, the program RAMs these are directly accessible by the MPU so a simple test shows them to be bad. So I cut all the legs off and remove the chips. Fit sockets and new RAM. Program ram is now ok.
For the vector RAM, they're still directly accessible by the MPU but now they're also MUX'd together with the vector stack / program counter. This time, as well as memory errors I'm getting address line errors (address lines tied). So my suspicion is the LS157's that are multiplexing the two addressing mechanisms. Sure enough, one of the LS157's is bad so I replace it.
At this point I'm now getting just plain old RAM errors. With the Fluke I can tell that one of the 6 x 2114's is bad as when I piggyback a good one on top the error goes away so that one is cut out. Next though I'm getting errors that don't go away with piggybacking a good chip so at this point I'm suspecting the faulty chip is holding a data line in a state that excludes that method of testing. Little choice at this point but to unsolder the remaining RAM, fit sockets and test them individually. Which I didn't do last night as it was 2am by this point and de-solder stations are noisy!
So really, my point is that even when you have all the tech kit in the world that allows you to pinpoint things properly desoldering and removing is often still the only sensible option when more than one device is powered up on the data lines at once.
Obviously when things are working properly a bad ram shouldn't affect the others. In my case there are 6 2114's. Each 4 bit data. So A1, B1, C1 are the high bits for 2000-23FF, 2400-27FF & 2800-2BFF respectively, A2, B2, C2 are the low bits for the same. When A2 was bad it only affected it's own memory space. But one of either B1 or C1 is also bad with a fault that is holding one or more of the data lines high. Therefore I won't know which it is until I remove them, one at a time, or just both at once and fit sockets. Even snipping the power pins for the chips so they're not powered up won't help if they have dead shorts in them holding one or more data lines either low or high.
Thankfully, the scenario I've described isn't too common. But sometimes desoldering is the only way!