Your results suggest that Cubase 9 has even more problems with this issue than 8.5. Maybe you should repeat the test with 8.5 to see if something else caused the difference in our results. 
The 48.4ms delay between your Tracks 9 and 11 suggest that your audio interface reports a 48.4ms output latency at that sample rate and buffer size - is that correct? The relatively short delay between your Tracks 11 and 12 suggest that your VSTi was inappropriately triggered immediately by Cubase when it detected the MIDI input. It should have been delayed by the reported input latency of the audio interface like the MIDI note recording was (Track 9) and my Track 6s were (Steinberg Forums and https://www.steinberg.net/forums/download/file.php?id=19479). Your VSTi audio (Track 12) precedes your mic audio (Track 2) by your audio interface’s input latency (48.4ms)
The 18.4ms delay between your Tracks 2 and 6 suggest that, unlike Cubase 8.5, Cubase 9 has additional problems when ALC is OFF. Your VSTi audio (Track 6) precedes your mic audio (Track 2) by that 18.4ms
Out of curiosity, what audio interface and MIDI controller did you use?
From page 1103 of the Cubase 9 Operation Manual
“Delay Compensation Threshold (for Recording) Cubase features full delay compensation - any delay inherent in the VST plug-ins you use will automatically be compensated for during playback." Should that be corrected too https://www.steinberg.net/forums/viewtopic.php?f=253&t=113330
It actually delays that MIDI recording by sum of the audio interface’s reported input and output latencies (reported RTL). Why is it delayed by reported output latency? What is that supposed to do
If it was it could do something worthwhile
Do you think you “adjusted” yourself better because you had 48.4ms less latency with ALC ON than with ALC OFF? Isn’t ALC ON really ALC OFF
If it worked properly it could be useful even at low latencies. Hopefully you’re not saying it’s obsolete and should be removed because it’s too hard to fix
So when “Adjust for Record latency” is active, plugin delay is included in record latency compensation and is not included when inactive? When inactive, record latency compensation is determined solely by the audio interface reported RTL and selected ASIO Guard delay. When active, the track with the largest plugin delay determines how much extra delay gets added to the compensation – right? Pro Tools conveniently shows the individual track delays - can Cubase
Wouldn’t “Include Plugin Delay in Record Latency Compensation” be a better name than "Adjust for Record latency” for this option?
I would like to have the option of preventing Cubase from adding VSTi latency. IMHO disabling MIDI “ASIO Latency Compensation” on the appropriate track(s) should do that.
What translation software are you using? (from/to German?)