Updated the above post.
That's an interesting way to look at finding the tiers for the visual GCD/Casts. One thing to consider though is that it doesn't work for Empyreal Arrow (66/15 = 4.4).
The minimum recurring pattern being 7,7,6 7,7,6 7,6,7,6 tells us that there are a maximum of 10 rounding points per 66 points of SS. So you would still have that same pattern given a 10 second Cast.
--------------------------------------------
What I propose is that there are only Tiers for showing the visual representation of SS. But the actual effect of SS is valued per point.
-------------------------------------------
Take for example the 2 second cast time column.
SS value 467 (+113 SS)
This data point is special because you have a decrease in Recast Time but not a decrease in Cast time. It lands on the recurring 7 data point for EA @ 14.74 Recast.
Data Point = 1.96/2.45
Calculated value of that Data Point = 1.712% (113/66) (98.2879)
2s = 1.965758
2.5s = 2.4572
-----------------------------------
The previous Data Point @ 460 (+106SS)
1.96/2.46
Calculated value of that Data Point = 1.606% (106/66)(98.394)
2s = 1.96788
2.5s = 2.45985
-----------------------------------
For whatever reason SE seems to have chosen to do their rounding @ the 4th decimal. That is why 1.96788 is rounded to 1.968 but visually in game only shows to the second decimal 1.96. But 2.45985 goes to 2.46.
Same with looking at the very first rounding point @ 361 SS
1 second cast @ +7 SS (calculated 0.106%)(99.893)
In game 0.99s
Calculated 0.99893
Visually in game this gets rounded to 0.99 when really it's 0.999 but you only see to the 2nd decimal place, but also the reason "why" it simply doesn't show it still as a 1s cast.
----------------------------------
So I think there are two things going on here.
Visually in game for how the Cast/Recast/GCD is displayed goes off the 776 pattern at which point it does a rounding check to visually display the effects of the stat.
The actual effects of the stat on skills is given for every point of stat - but is really hard to pinpoint due to:
Attack Speed Swing (i.e. 2.5s GCD could be in the range of 2.45-2.55) so it takes a lot of averages to get accurate results.
Server Latency
System Latency
Perfect Testing - Requires perfect skill queuing for 100's of data points.