Thanks for taking the time to explain that, Joe. I understand what the problem is now - where you have a many-to-one relationship between the two joined tables, the UPDATE FROM syntax effectively updates the parent table with all of the values from the child table, leaving the value from the physically last row as the one that persists.
I'm still not 100% convinced by your portability argument. I absolutely agree that just as a writer should know standard English, a developer should be aware of the standards in any particular programming language. However, programmers (and writers) should also know when it is appropriate to deviate from those standards. Each DBMS has its own features that distinguish it from its rivals and which, in some cases, provide extra features or performance. I see no reason why such features should not be used where they improve on ISO SQL.
I also agree that maintenance may be more difficult on non-standard code. When dialect provides no benefits above what can be achieved using standard code, you are right: the ISO-compatible construction should be used. This is not a reason, though, to fail to take advantage of the features of any particular DBMS. Needless dialect should indeed be avoided, but because it's needless, not because it's dialect.