Comment by WalterBright

The CTFE interpreter knows what "new" does, and just calls the compiler's "new" function to allocate memory. If the user wrote their own custom allocator, then the CTFE would have to interpret that, which is kind of a big mess.