Currently, tests with many network events are very slow due to the test design and not to consensus being slow:
With many network events, and each node voting for them in a random order, it may take a long time before a supermajority of nodes vote for the same network event and the consensus protocol even starts.
This makes it look like PARSEC was slow, but really it’s only the tests being unrealistic.
We probably will want to order the events and vote them in that order + some amount of random fuzz to mimic a realistic scenario better.
It’s agreed that the extent of the randomness shall be adjustable by a factor so we can figure out which one will be more realistic.