Simple methods like Peter's are usefull and can be tested easily.
If undesirable side-effects are detected, one can be sure they will appear on this or related fora soon. But, until a practical case is shown that disproves the usability it should be tested...
Unlike in mathematics, in information science only very simple algorithms/procedures can be proven to be correct;
know your Knuth ! On a logical level only algorithms like Dijkstra's ... Certainly not any of the procedures given in discussion fora ...
The source of the Delphi forum is a renowned veteran user, who certainly would not present a method he wouldn't have tested long ago and exhaustively.
On a low level (translation to assembler, and even more, the dependency on API-functions of the OS) different programming languages on the same platform are still quite similar. So if the procedure "works" in one language, it it quite likely to work
in other situations to, if adapted. But a counterexample could prove this to be false of course.
You are certainly right that the method, as presented, is not cross-platform. But that doesn't mean it should be (theoretically) impossible to adapt it to other platforms, as there is a similarity in underlying mechanisms of message-processing.
The question is more: is the need there, or is it's manifestation (of problems) different?
(side note: I tested a few small programs that, in Windows, without ProcessPaintRequests, had some problems with updating the UI. But in my trusted OS/2 4.52 -version inside VirtualBox, they had no problem if recompiled for OS/2 there. Go figure ...)