Consider the case of converting an integer value to floating point. In order to normalize the floating point value, you'll have to do something equivalent to counting the number of leading zeros on the int, and many machines don't have an instruction for that, so you have to do a loop. Data-dependent timing like that needs to be studiously avoided in crypto code in order to prevent timing attacks. (Of course, crypto rarely uses floating point, but it's still a great example of how an O(1) operation can have variable execution time depending on the input.)
I wonder, how do you propose to measure the "size" of your input to bool cast?
I think that might be possible in dynamic languages like Lisp that can compile to assembly. Maybe Franz or Allegro Lisp compilers do something like that. For the JVM or .NET, I'm not sure that can happen. For C or C++, every typecast is constant time.
And that might differ between JITs and compiled languages and different numerical types.
For any machine precision numerical types, if the code is compiled, it'll just become
TEST EAX, EAX ...set eax to 0 or 1 based on jmp...
So in most compiled languages, a bool cast on machine precision numerics, will be constant time.