In fact it is written in the comment right above:
// NaN means non-codepoint
So it is mandatory to perform this comparison first to handle the NaN case as in JavaScript:
NaN === NaN returns false.
As pointed out by James Wiseman it is also important to know why the developer used high !== high instead of isNaN(high) which would have been clearer.
It is certainly based on performance. This test shows that a !== a is twenty times faster than isNaN(a).
zzzzBov also indicates that isNaN() could be overwritten, using !== is also more portable.
More info from Benjamin Gruenbaum:
It is also worth noting that NaN does not equal to anything else as
well, and also it is not equal to anything else in an unstrict sense
And from Jan Dvorak:
Also note
{valueOf:function(){return{}}}does equal itself