Let's see how the grid may affect the results on Taylor
The new grid was made with Gambit too, but using four times more cells, with a configuration essentially thought for radial and vertical gradients.
New grid with 50000 cells
This grid refinement allowed to catch the instabilities for a Taylor about 1700, which is near the theoritical value of 1712:
So the lenght of cells seems to influence the apparition of the instabilities.