That was absolutely true way back when the industry was new and there were myriad system vendors with their own operating systems.
We have learned a lot since then, and we know how to address these problems: openly specified document formats and protocols. Programming languages with portable runtimes and libraries. Open source software. Web Applications.
The world is a much different place from when you felt trapped because your Visicalc files were on LDOS floppies that could only be read by a TRS-80 Model IV.
Not necessarily. Web standards could be a great foundation. The web is open and anyone is free to push the state of the art forward. Yet the result has often been more interoperability, not less.
Yes. Not in the way they are right now, but yes, web based applications, running off remote servers, maybe interfacing with a desktop client or something similar to that is the future.
Some desktop apps yes, actually many of them have been already replaced. Other (data/processor intensive: graphic arts, multimedia, etc...) don't think so.
I think the benefits of having a diverse OS landscape are much greater than the cost of a little increased complexity. We already have languages that run on virtual machines, and we have the POSIX standard. Software portability shouldn't be a problem.
Haiku actually does this. Haiku is not a UNIX, but there's a fairly complete POSIX layer for building UNIX programs. As far as I know, Syllable and SkyOS do something like this, too (I might be wrong, though).
I agree. Ask someone who was involved with developing Borland's Turbo Pascal. When I used to help duplicate it (think digital Kinko's) we had what seemed like 50 masters to dupe from: DEC Rainbow, Heath/Zenith, HP, ACT Apricot, NorthStar, CP/M, Eagle, Pixy, many, many others. Oh, and MS-DOS and IBM DOS.
It would also be a nightmare of complexity.