Personally I prefer to go unbiased wherever that is possible. For that reason, I don’t like VCM very much. However, there is an unbiased version of it that I like.
This together with the above pixar paper and, perhaps, the improved MIS that can be found here https://graphics.cg.uni-saarland.de/publications/kondapaneni-2019-siggraph-optimal-mis.html would currently be my ideal scenario, I think.
Assuming all that can be made to work on the GPU. 'cause the example scene they show surely has complex lighting and so perhaps warrants the 1h CPU rendering. But on GPU it should presumably go rather quickly.
(I’m also not sure how all these techniques play with volume rendering which is also, imo, in need of an unbiased overhaul)