Since there are problems with values greater than 0, I have tested the playback of stereoscopic still images with pan / crop.
Here are a few values, which I have determined.

GPU + 0MB - 2.2fps
GPU + 100MB - 5.5fps
CPU + 0MB - 1.5fps
CPU + 100MB - 4.5fps

Larger values than 100 MB (200 MB default) bring no advantages here, possibly with video files. They are only needed for dynamic pre-rendering.

Dynamic RAM 0 -> preview very sluggish but correct
Dynamic RAM + 100MB and GPU on -> flickering, wrong frames, black frames, crashes

The Multicam Edit may also be affected by this issue?

CPU + 100MB -> preview correct

looks like ZERO is still a hero

This will depend on the specific machine and on the type of footage. Both with XAVC-I FS7 and EVA1 long-GOP footage I see on my system a better playback if the dynamic ram preview is not zero but somewhere about 100 to 200 MB.

For a good playback rate, I work with the default values. If I need the RAM cache for pre-rendering, it can be more than 4GB. Only if the issues described above occur I tentatively set the value to zero. Since no restart is necessary, you can play with the value.

True. I constantly adjust this in a session depending on need.

Why does that setting affect anything in the first place? Does not make sense to me.

Mindmatter, I kind of agree. I heard about it in Vegas 12 and it made a HUGE difference back then. If you do SHIFT-B to "render a selection", that memory number uses ram to render it for playback, I guess. I don't know why ZERO would allow timeline playback to be faster. It's just old, old coding.

RAM cache is generally used to speed up playback, in addition to the pre-rendering with Shift-B described in the Help. The test results and the feeling of the timeline playback show it. The problem is "only" that sometimes something goes wrong with gpu usage.
In addition, the cache usage does not automatically bring about an acceleration, in this case a value zero would also be appropriate. Source material, hardware, operating system affect the result.

We won't know until you run the tests

So you do your own tests with a project that is slow enough to capture the differences.

That RAM assigned in the dynamic RAM preview setting also buffers frames when part of the timeline is repeatedly playbacked, e.g. in a loop playback (without using the RAM Preview function, of course).

When I put a 24 fps HD video of about 10 seconds length into the timeline, apply an FX which eats some system resources (tested with Neat Video) and RAM is set to "0", the playback is around 5 fps and won't get higher when looping.
When RAM is set to default "200", after several loops the playback performance rises up to 6,5 fps.
And when RAM is set to "4096" (my personal default), after several loops the playback performance rises to 24 fps (full speed).

It also affects render speed but it doesn't always mean in a positive way. While rendering that same piece to XDCAM EX with RAM set to "0", it takes 2:29 here, RAM set to "200" renders in 1:55, and rising the RAM level above 200 MB does not speed up rendering anymore.