Depends. It's a cost-benefit tradeoff. If the use of floats is appropriate, 53 bits of precision is probably way more than enough.

By the way, Perl natively supports (without loss of precision) integer numbers (as in numbers without a fractional component) up to 2**53 (or higher depending on platform and build options), not 2**32-1.