I am not putting this topic into bug because this probably isn't.
Is anyone using this in production ?
I did about 50 tests today and same result, with simplest Shadowcatcher (everything on default), composite mode takes 10 times longer to clear.
Nothing is wrong in stats, same ray/s, samples per pixel, total rays. Same. Except for noise. The image starts from 20+ perc. instead of 6 with any other mode. I tried adaptivity on/off but that is hard to judge since noise ratio is not being shown in latter.
Is it because of the forced black ? I can't wrap my head around why it would need so much more sampling. The shadow alpha is incredibly useful, and the two other modes are totally useless.
For workaround, I am using just the non-compositing modes, and select the whole area by regular alpha and tweak it with levels. Only works because I need just white sweep..