The point of Taos was not network transparency, or clusters, or any of that.

The USP of Taos was platform-neutral binaries, translated on the fly to native format as they were loaded from disk. Full native performance on every platform from a single binary file, with the loader/translator embedded into the OS itself.

[Aside: It ended up very small compared to native OSes so it could be used as a runtime as well. It was so efficient that it could convert Java bytecode to its own format and convert that to native code and still outperform the JVM... for a while. ]

Something akin to Java, but the JVM is an addon that you install on a different OS, and there's no "native" OS*. Apps aren't native and don't look or feel like native.

Plan 9 has nothing like this; Plan 9 is UNIX 2.0 with networking and GUI right there in the kernel. Binaries are written in C and compiled to the native CPU; no portability.

Amoeba OS I've never seen but I think it was much the same.

Inferno is Plan 9 2.0, which is to say, it is UNIX 3.0. C is banished, native binaries are banished. A runtime VM, Dis, is embedded in the kernel. Binaries are written in Limbo and compiled to Dis. The same binary runs everywhere, like Taos, but it's not translated as such: there is a bytecode interpreter at a low level.

[Aside #2: It ended up very small compared to native OSes so it could be used as a runtime as well. It was so efficient that it was competitive with the JVM, and so was offered as a rival, but still with its own format that could outperform the JVM... for a while. ]

[Aside #3:

Note the very similar asides 1 & 2. That's deliberate. Both fell into the same trap. Both failed and I think this was a contributing factor. ]

[Aside #4: Java fell into the same trap and failed to compete with Javascript embedded right into the HTML renderer itself. ]

-----

* Yes I know about SavaJE etc. Not directly relevant, and all flopped.

Having said that, this seems like a natural evolutionary path for the Boxes OS.