How Reliable Are Browser-Based GPU Benchmarks?

I’ve been comparing traditional desktop GPU benchmarks with browser-based tools and wondering how accurate they really are. Since modern browsers support WebGL and real-time shader rendering, they can apply meaningful stress to the GPU under sustained workloads. I recently tried a volume shader tool and found it interesting how frame stability differed over time, you can Run the test directly in the browser without installing anything.

Do you think browser-based benchmarks are reliable enough for serious comparison, or are they mainly useful for quick checks?