Didn't use lockscreens as the main thing that bothered me was the 6.7/6.5 disparity (which activated my spidey senses). I did test a bit with lockscreen yesterday too, but it was good that I didn't too much since that resulted in me fixing and improving the 2 tests.
The results of running again (all numbers are milliseconds)
the stack script (icons): https://pastebin.com/nkq83aed
This time I added a part to remove the duration of the repeats and other operations, which didn't add much, but more than nothing. Changed from ticks and seconds to just milliseconds since using the other 2 was ... weird. Also instead of arbitrary 120 graphics and 2000(*2) icon swaps, the final numbers are the calculated averages for a single operation.
For the lines/points swapping,
6.0.2 to 6.1.3 line's point swapping really was abysmal as it looked like before, there was also a massive improvement in 6.5 that was slightly negated in 6.6.2 to 8.1.7 (almost 4x slower) and >9.0 normal again.
Interestingly the benefits of locking the screen seem to have become consistently less than half as beneficial. Still the a difference between locking and not is notable (~20x improvement in 9.0.1), but at this rate in 2034, it will be pointless to use lock screen
As for the icons, the end result for not locked is what I had yesterday and with lockscreen the times were fairly stable across the versions. Difference being though is that prior to 6.5, icon swapping was almost as fast with or without lockscreen and there was no improvement to the time it took in newer versions, only the unlocked icon swap got slower making the benefits of lockscreen more apparent. Which I think is what caused tjframe's conclusion.
The general impression I had gotten from reading these forums whenever I had a problem was that the standalone would run at mach 10 while the IDE strolls at walking speeds. Not the sub 1/6th benefit from my tests.
It's great if all you use is LiveCode, but if you already use dozen other languages each with a dozen other libraries and frameworks then decrypting LCB to a point of usability isn't quite worth it, yet. That's just me though.