The WS2812B cycle time is 150 ns.
WS2812B communication is done at 800 kHz, so the cycle time is 1250 ns.
The distinction between 0 and 1 is done by varying the time the signal is high within those 1250 nanoseconds. On the tighter timing end, 20% low is sometimes used, which is 250 ns. So you'd want a driver chip with specs significantly faster than 250 ns.
74HCT245 is much better than 90 ns, which isn't even the spec that truly matters, as I'll explain in a moment...
Most datasheets for 74HCT245 also cover 74HC245, which makes finding the right specs more difficult than it should be. Here's the parts of one of those datasheets covering 74HCT245.
74HCT245 is plenty fast enough for WS2812B use.
The 74HC245 chip is also fast enough, when used at 5V. But it's rated for a wider range of power supply voltage, from 2V to 6V. They give specs for a variety of different voltages, which makes things more complicated.
The slow 90ns spec only applies for use with 2V power. The "Conditions" column is the key.
When run with 4.5V, the 74HC245 is also plenty fast enough for WS2812B use, even slightly faster than 74HCT245. However, elsewhere in the datasheet says 74HC245 is only guaranteed to recognize inputs over 70% of the power supply as logic high. That's why 74HCT245 is normally used. The extra "T" in the part number means it recognizes inputs over 2.0V as logic high, when powered from 5V. When your signal is 3.3V from Teensy, you want the HCT part.
The thing which
really matters, and also isn't very clear in these specs, is any difference between low-to-high versus high-to-low. Since WS2812B relies on pulse widths, if the performance is mismatched for the rising vs falling edge, then the waveform's pulse width is changed. A relatively long propagation delay is actually not very bad, as long as the same signal which went in ultimately comes out. While not explicitly mentioned in these specs, chips like 74HCT245 and 74HC245 are very good in this regard. They can't be any worse than the minimum delay for one case (which is zero) and the max delay for the other, going only by the specs, but in practice the rising and falling cases are usually pretty close to each other.
However, the bidirectional chips are usually terrible in this regard. The rising vs falling edge performance is often mismatched by 10X or more. They're usually fast at high-to-low transitions, but slow low-to-high changes. The output impedance (or drive strength) is also vastly different for high vs low output. Some bidirectionals chips have a one-shot timer for a low impedance pullup, which helps, but still the high and low performance are quite different and the high output is very weak after the timer finishes its short pullup pulse.
These are the reasons why unidirectional chips like 74HC245 should be used. Not only are they fast, but the rising & falling edged speed is well matched for a low-distortion signal. The output drive impedance is consistent. Those are the properties you want when trying to send your signal down a wire to your addressable LEDs.
And when you run these chips at 5V, their specs are plenty fast enough. The really slow specs only apply when using 74HC245 with 2V power, which you wouldn't do for driving WS2812B LEDs that want 5V signals.
Hopefully this long-winded post clears up confusion about the specs. Reading datasheets is hard and understanding which specs truly matter is even harder, especially when none of the datasheets gives you an explicit mention of the specs you need.