The other reason why we didn't invest more into porting the Native Image serial GC (or G1) is that the WasmGC proposal was coming up anyway, and JS runtimes anyway already spent a lot of effort into optimizing their GCs.

The two main reasons why we didn't try to optimize a native GC is because packaging it in the application image would make the binary much larger, and we expected that running it in managed code would be slower than in native. So we initially made a custom small GC implementation for linear memory.