for (1,2,3) { }

but not here

 for (1 .. 3) { }

In the first case Perl (historically) tries to avoid the overhead to allocate a variable by pointing to the already existing literal in source code.

At least thats what's happening in (e)Lisp, one of the languages which influenced Larry.

Literals can only be constant since the source shouldn't be changed.

The operators x and .. OTOH calculate the values, there are no source positions which can be used for optimization.

I don't know if this optimization still makes sense, but I think it's at least the historical motivation.