So in practice it solves 99.99% of the problem
by "it", do you mean the timeclock?
Yes. A semi-reliable time allows a bunch of entities to collectively decide that another entity is "dead". This makes non-Byzantine distributed consensus easy (by non-Byzantine, I mean it handles "normal" failures -- node death, network split -- but not pathological cases where a node is deliberately trying to break the election). This is all that is needed for distributed consensus in most tightly coupled distributed systems.
Very interesting post.
Question: would the problem of the time interval case you describe (with an imprecise time signal) possibly be solvable by having the clients decide on a time to accept the block in the form of a probability distribution, peaking around the agreed upon time point? that way, assuming a majority of clients would use that rule, we could assume a majority of clients would accept the block issued *around* time E, even if it is not issued at the *exact* time.