I am opting to answer on "why the second approach here is working".
In one of the WWDC videos published since 2010, they said that :
+[NSImage imageNamed:] chooses the best image representation object available for the current display.
So chances are that you are calling this class method from within a locked focus context (e.g. within a drawRect: method or similar), or maybe you actually called lockFocus yourself. Anyway, the result is that you get the most suitable image. But only when calling +[NSImage imageNamed:].
EDIT: Found it here: http://adcdownload.apple.com//wwdc_2012/wwdc_2012_session_pdfs/session_213__introduction_to_high_resolution_on_os_x.pdf
Search for the keyword "best" in the slides: "NSImage automatically chooses best representation […]".
So, your second version will return the Retina version on a Retina display, you can be certain of it, it is advertised in the documentation[*].
[*] This will only work if you provide valid artwork.