And the next round about 15 years ago -- for Croquet -- which included Dave Reed, David A Smith, Andreas Raab, myself, etc. There were three or four different versions of this mechanism, all of which worked, and the designs progressively got better.
There are several sets of issues and tradeoffs. Reed's initial ideas in the 70s were about what an operating system for the whole Internet should look like. One of the many interesting ideas he proposed was to deal with the slow latencies and large numbers of potential users by distributing cloned computations organized by pseudo-time, and using the slow Internet for just input and synch deltas. This is what we first implemented in Croquet in the early 2000s when a typical good Internet round trip for a ping was about 80-100 milliseconds. This is good enough to create a massively parallel game like World of Warcraft (or even a flight simulator game) without any servers at all by just using the distributed machines that the game players are using to play on.
Further versions did more, and more comprehensively.
(A side note is that Croquet without the real-time graphics and interaction, etc. automatically provides a distributed object-oriented data base, etc. Pseudo-time is the larger idea underneath protected distributed atomic transaction data-bases, etc.)
Your last question is a good one, and it has been discussed over the years. Can it be done? How much would need to be done, and where? Etc.
The object work we did at Parc was a part of the network thinking going on in the ARPA/Parc community. And there was an eventual intent to think of higher-level network entities. A quite good intermediate real implementation of "net level entities" was the LOCUS system by Gerry Popek and his group at UCLA after he spent a year at Parc thinking about what network systems should be like (this was really good, and the first several chapters of his book are well worth reading). It was a migratory load balancing idea over heterogeneous machine types, which was quite complementary to the pseudo-time ideas.
I would like to see a highly talented group take another pass at your last question. Resilience to scaling is quite difficult to predict ahead of time, and it's good to do real implementations. Still, the 9 orders of magnitude that the Internet needed was possible with the scheme that the ARPA/Parc community had, and Croquet worked perhaps better than it should have (especially with just a few people working on it).
In any case, I think it is a bit crazy to not use pseudo-time if you know about it, and it's a bit amateurish not to be aware of it. In the middle are religions whose identities are built on various "one true way"s (and this is the opposite of the way any science should be conducted).