As yenyen notes in the comments, the issue was that the input was UCS-2 (probably really UTF-16) encoded.
UCS-2 is a two-byte-per-character encoding that contains null bytes. If you tell PostgreSQL the file is utf-8 then it'll see the input as garbage full of invalid utf-8 sequences. If you tell PostgreSQL it's a simple 1-byte encoding like latin1, PostgreSQL will see the zero (null) byte and realise it's not latin-1 after all.
The trick here is to examine the input file with an editor that can show the raw bytes, not just use a text editor that automagically reads the BOM and loads it as encoded text. If in doubt use a hex editor.