https://d3i71xaburhd42.cloudfront.net/605efbc0fed86899dcbf87...
Christine Flood is pretty great and she has some wonderful talks on YouTube if you're interested in how Shenandoah works. The image above comes from one of her papers on GC (https://www.semanticscholar.org/paper/Shenandoah%3A-An-open-...) and shows order-of-magnitude improvements in pause times with average pause times 1/10th that of G1 and total pause times 1/27th that of G1.
Shenandoah and ZGC are also made for larger heaps and often datastore have larger heaps.
No doubt we could have spent time optimising this but moving to G1 brought this down to rarely exceeding 100ms, Shenandoah brought this down further to typically <10ms.
As far as latency goes the new collectors do perform better in my experience, but you should try and measure.
But yes, they also affect mean latency.
> The period of time during which the garbage collector is active [and threads are stopped] is referred to as its latency. [1]
This does of course affects the distribution of response latency. High GC latency causes more varying response times (though not necessarily higher average response time).
[1] https://docs.microsoft.com/en-us/dotnet/standard/garbage-col...
Never got around to trying any of the newer GCs before it was rearchitected.