OK, here's everything I found with my clutch testing rig, some of which I found a bit surprising.
I tested a dozen or so different oils - motorcycle oils, ATFs, non-friction modified engine oils, diesel engine oil, CVT oil, hydraulic oil. Nearly all the oils gave very nearly the same results; the CVT and ATF oils were at the bottom of the list, followed by pretty much all the others apart from the motorcycle engine oil which had a clear lead over everything else. The one I tested was a Castrol synthetic 4T oil; I expect other brands would give similar results.
In general, higher viscosity oils held better than low viscosity oils - something I wasn't expecting. This was particularly so with plates that had a smaller surface area. The exception was the worn factory plates that didn't like the thick oils. The obvious tradeoff with viscosity is cold drag.
I tested several types of plates - new factory plates, worn factory plates, Barnetts, Honda style plates with cork facing, Honda style plates with cellulose facing, plain steel plates I had laser cut in two different thicknesses.
The factory Bul plates appear to have a massive area, but they are embossed at the factory in a process that leaves hundreds of little peaks on them and this reduces the effective area by a lot. Out of all of the plates the new Bul plates were second lowest in holding power.
The worn factory plates gave the lowest hold. As the little peaks wear down the area goes way up, and it appears they then hold a thick film of oil between the plates that reduces the friction very significantly. Worn plates simply will not hold.
The Barnetts use a Kevlar facing that seems to work well - they gave more holding power than the new factory plates. These plates work well in lower revving engines but are problematic as the rpms and power rises. To keep the plate thickness similar to stock the facings are bonded to a very thin backing plate, and I found with my air-cooled engine I could kill the Barnetts in an hour or two on the dyno. The thin drive ears would be badly beaten up.
The Japanese style plates were better again. They have a larger inside diameter and a small contact band. The cellulose facing gave a slightly better hold than the cork but there wasn't much in it. On the dyno however something strange happened - the cellulose plates worked perfectly for a while but then suddenly lost their grip. There was no visible fault with them, they just went slick for some reason. I switched to the cork facing and these ran without problems. The thicker aluminum backing held up to the torsional battering fine. The combination I ended up using was cork faced plates (modified Kawasaki plates from ProX) with thinner steels I had cut out. I use this combo in all my engines, however if they later prove to be too weak for the LSR engine I'll go to all steel.
Plate thickness is important - obviously the thinner the plates the more of them you can run. At first I thought that going too thin would be problematic and the drive ears would get beaten up. But that proved not to be the case - regardless of plate thickness the drive ear total thickness remains the same and even with 1mm steel plates the ears survived just fine.
The strongest grip by a large margin was given by the custom steel plates. These had a very narrow contact face and a large effective radius. In actual riding they tend to be a bit savage on the take-up but it is something you can adapt to. Even with the stock thickness they gave a big improvement and with the thinner plates they are even stronger. I think these could handle 100hp, should I ever get to that figure (highly unlikely). They are very cheap to get laser or waterjet cut but you need to be careful to use dissimilar grades of steel for the inners and outers. My initial set used plain mild steel for both and while they tested fine on the rig they would gall and bite in actual riding. Nickel plating one set of plates fixed this, at least until the plating wore out. One set of plates need to be made from a tough manganese steel or similar (Waukesha 88?) to avoid this. The ProX plates are working for me at the moment and have a nice smooth take-up, but if need be I'll switch to an all-steel pack.
All these tests were done on a Pursang style clutch which is smaller than the Bandido unit but uses the same type of factory plate, so everything should apply equally. One key difference though is the chain drive setup on the Pursang makes it easy to change the primary drive ratio to reduce the torque loading on the clutch and gearbox.