The only way to know this is to run a car on a chassis dyno, then take a engine out of a car, put it on a engine dyno, and record the power output. But then you have another problem. Both dynos have been calibrated to different specifications with no COMMON reference. I have never seen an example of this being done. The car manufacturers may do this but nobody else would spend the money to do it aside from professional race teams like NASCAR for example. Those guys would never divulge that information.
Dynos are really a tuning tool that is used to compare output before and after tuning adjustments.
Considering the huge difference in chassis dyno power results between dynos I really have no idea how much difference there would be. But my completely wild guess is about 50 HP. I think it would be a fixed power loss for a given car, not a percentage of crankshaft power. Ya I know it can be calculated but I also don't believe that if you doubled the engine power output the losses would also double if you left everything as before the power increase.
My GT40 has a 345-50HP 302/R21 Renault in it and it made 296HP on a chassis dyno. My SLC has a 450hp 350 SBC G50 Porsche in it. It feels like it has about another 100Hp than my GT40 has.
If you Google this question...............well it's the internet............numbers are all over the map. 50-200HP and 2-25%. By the way, do you really think a 10,000 HP top fuel car has 20% losses in the power train? That's 2000HP or well over a MEGAWATT ! If you sink that much heat into the few quarts of oil in the differential it would flash into fire and the ring and pinion would fuse together. It can't be that much.