Damnit! My server crashed two weeks ago and I had similar data that both supports and contradicts your findings.
I'm sure this has been done plenty of times before but I didn't couldn't remember where that research was so I just did my own.
That was me (At least one of them, I'm sure there were others). I did a bunch of tests on FC/SST/Nightingale as well as HP/Latent. I also supported the findings on BG regarding the specific timing on equip macros and found that sometimes the server doesn't always swap gear fast enough (this was later confirmed with spellcast btw).
Unfortunately, I'm breaking a cardinal rule when it comes to talking about game dynamics and proving anything - Namely: Evidence. So if you must, take it with a grain of salt - Maybe simple discussion will re-motivate me to perform my tests again.
First: All I can say is that the start/stop timing is important. Usually when I read about these types of tests, people try to eyeball it or try to base it off the casting bar. If your talking about time, you need to find references that always happen at the same moment every single time, and you need to refer to those test in measures of time (such as secs, or something convertible such as frames). Fortunately, you did just that.
The first frame when you sing without an instrument as a taru, your eyes close and your mouth opens, so I started on that frame.
This sounds reasonable if you can catch this every time (ie: no half-eyes closed or mouth part opened). I used a different method - When you have nothing selected and you 'start' a song, a cursor will appear immediately over your head. The cursor doesn't fade in, its either there or not - and was a good indication.
Using the yarnball plugin, I picked the exact frame the buff duration timer appeared on the screen as the last frame to count how many frames the song took to cast.
Again, sounds reasonable - I can't even dispute this one due to the fact there's likely no half-way or fade-in. I used the same method previously however - With nothing selected, when you complete a song a cursor will hover over your head the moment the song finishes.
The only thing I will say is - Your frames vary quite a bit compared to mine. Maybe it was the eye/mouth part, or maybe something else, but when I performed tests I always had +-1 frame difference. The great thing was you could add up the frames off my naked tests and calculate the times, and it was always 8 seconds and/or +-1 frame (1/30 sec). At least in this aspect, my tests were more accurate than yours (but whats a few frames?).
I also performed my tests in a non-lag area (the chocobo racing area is perfect). Unfortunately my goal wasn't to find different casting times in different gear, but rather to see how FC/-SST/Nightingale stacked (Inconclusive, I couldn't find any correlation - I'd wear one setup and it would work one way, I'd try another and it wouldn't be what I expected). I know I came close to 50% with gear though.
What I will say is that 50% (4 secs) is NOT the absolute cap - Nightingale WILL reduce casting times further. IIRC I believe Nightingale always took place first and separate from the rest. Songs can be reduced to just slightly over 2 seconds (I could never get an exact 120 frames).
I will say though that the tests I performed were before the addition of D.Knifes, ACP Bodies, AMK Hats, and Ebon Talar sets - It's always possible SE changed something.
I would be really interested in seeing frame data for each individual piece of new gear though.