Cirno's attacks are also faster than average, don't forget to account for that (I did at first, and it does matter).
Explanation that abstracts things a bit: every attack in the game has a delay value, which is how much is subtracted from the CTB meter when the attack is used. The rate at which the CTB meter fills is based on Speed, and LoT PC speed has stepwise diminishing returns at values above 200 (speed and charge rate have a 1:1 relationship up to 200 speed, then a 2:1 relationship from 201 to 300, then 3:1 from 301 to 400, and so on). (LoT enemy speed does not have diminishing returns, and is always a 1:1 relation to CTB fill rate, so halving an enemy's speed makes their CTB take twice as long to fill.) In practice, the net DL effect is not quite exactly half speed because of debuff decay; the debuff decays to -40% when Justin takes his turn, so until Cirno gets her next turn, he's filling CTB at that -40% rate, not -50%.
Every tick, all combatants' CTB is incremented by an amount based on speed (the increment amount is shown in the stat topic in brackets; for Cirno, it's 320), and when it reaches 10000 (or higher), the combatant gets to act. Excess CTB over 10000 is lost, but is used for tiebreaks. LoT average speed by the existing stat topic is 385, which works out to 278 CTB points per tick. LoT PCs usually begin fights at 5000 CTB, so the average LoT PC gets their first turn after 18 ticks. Cirno gets her first turn in 16 ticks. (There are a few exceptions in boss fights IIRC)
The average attack delay given in the stat topic is 5934; thus, after using such an attack, the character's CTB meter will be at 4056 and will refill from there until it reaches 10000 again. For an average speed PC, this happens in 22 ticks (21.34, have to round up given the nature of the system). Icicle Fall and White Album both have delays of 5000, faster than the 5934 average; so after using them, Cirno only has to refill from 5000 (so she acts every 16 ticks).
Therefore Cirno gets her first turn 11.1% faster than average, and her subsequent turns are 27.27% faster than average (with the attacks being used in this fight; Perfect Freeze is notably slower than average at 7500 and Diamond Blizzard is basically dead average at 6000). [This is why Icicle Fall is her best damage despite being less damage per hit than Perfect Freeze.]
I'm frankly not sure how whoever wrote the LoT topic came up with the abstractions for speed debuffing (the 66% figure), but based on my own in-game experience it's closer to actually being 50% than that (but is not actually quite 50% in practice for the reasons stated above. 55% might be a decent abstraction, the true value depends a lot on the speed differential between the two combatants). My best guess is that the 66% figure assumes that the target is another LoT PC (in which case that would be a lot closer to right: LoT Speed debuffs are less effective against LoT PCs due to the diminishing effect). Interp issue, perhaps.
