Excluding -nan.0 was an oversight on my part, and it's gratuitously incompatible with R6RS as well as current practice. Racket, Gauche, Chicken, Guile, Chez, Ikarus, Larceny, Ypsilon, STklos all support nan.0 and +nan.0 as equivalent forms. MIT, Bigloo, Scheme48/scsh, SISC, SCM, Scheme 9 don't support either form. Only Gambit and Chibi support +nan.0 but not -nan.0.
STklos prints both +nan.0 and -nan.0 as -nan.0.
The reference to nan.0 above should be to +nan.0.
Updating list: Kawa, Mosh, IronScheme, KSi, VX support both forms. SigScheme, RScheme, Scheme 7, Elk, UMB, Oaklisp, Owl Lisp support neither. So it is still only Gambit and Chibi that support +nan.0 but not -nan.0.
The WG voted to adopt this proposal.
This is a pure lexical syntax change: there is only one NaN object.