I'll bite... their claim of 55ms is most likely for a tiny
region to be updated. It is not for the whole screen. You'll notice in their video that full page updates are still slow.
I've worked w/ e-ink technology and designed similar hardware before. Here is how they work:
You scan across the matrix, setting the row/column multiplexer to the dots you want and flip the dots between black and white as needed. This flip is a physical process and has an inherent latency. You can't move to the next pixel until you've finished flipping this one.
The flipping process is basically applying a high positive or negative voltage to the tiny plate under the pixel, which causes either the positive or negative side (black or white) of the embedded dot to show. The amount of voltage you apply determines the intensity. If you want crisp edges and accurate shades of gray then the voltage also depends on manufacturing characteristics of that specific panel. E-Ink actually gives you a waveform for each panel they hip, and that waveform is also temperature dependant -- because the liquid the dots are suspended in changes viscosity based on temp.
So you can do a bit of simple calculation and determine the voltage to go from say 25% gray to 55% gray. This is faster because normally you'd go to either black or white and then to your target shade. This is why you see that screen "flashing" when there is a full page change.
To bring it all back... if you only need to update a small region, say the area around the tip of the pixel, you can do so relatively quickly -- compared to the whole screen -- hence their 55ms claim.
Personally, I noticed a fair bit of latency in their video.
As an experiment for a previous project we rolled our own FPGA based driver and were able to get small region update down to about 20ms.