As a builder, I’m actually pretty impressed by the resilience of the system.
I’ve noticed that although some of my canisters have slowed significantly, calls are still (eventually) making their way through.
I can see through logging that one canister has already received messages back from the NNS subnet, and is currently patiently waiting for additional messages to reach their destination on the NNS subnet. Even though the finalization rate on the NNS subnet has dropped to 0.02 blocks/sec (1.2 blocks/min), messages are still making their way through.
On these canisters we didn’t end up with inconsistent state, unexpected errors or lock up - ICP being a state machine and having deterministic behavior really helps in circumstances like this.
Think about it this way. Everyone got used to just how fast ICP is (NNS subnet is usually 0.82 blocks/sec, most application subnets are usually ~1.3blocks/sec). Now we won’t take it for granted 
Also, I’m really excited for the Scalable Messaging Model to come out 