use physical sss.
You do understand thet fast sss is afake with capital F it asumes you dont rebounce light all that much, otherwise youd get a O(n^2) against O(nlog(n)) computation now offcourse gi itself is a O(nlogn) operation or possibly so the rersult woudl be a O(n^2*n2log(n2))operation, so quickly calculating you d push a magnitude one miunte render into 0.5-20 hour render its hard toc claculate exatchtu because teh ns are independet variables tough.
However theres a faster track wich is to do tis all by photon simulation but you need hellofalot of photons.
HOWEVER what i do suggest you dioo is make a phpotonic material for teh fast asss adnthen teh photons would contribute normaly but not sss totaly right but do a pseudogoodjob
Anyway turning all goodies on is performance scuicide.