What I *think* (after some testing) is going on is that the Math::BigInt and Math::Bigfloat overload the arithmetic operators, which breaks timegm().

No, overloading of operators is based only on the arguments given to them. But 'use bignum;' overloads numeric constants. So, for example:

    timegm(0,0,0,$day,$month,$year)
[download]

becomes

    timegm( Math::BigInt->new(0), Math::BigInt->new(0),
        Math::BigInt->new(0), $day, $month, $year )
[download]

And Math::BigInt might not be careful enough to have Math::BigInt->new(0) act enough like just plain 0 for timegm()'s usage (not surprisingly).

Is this expected behavior and/or documented anywhere?

It sounds like you probably read the first sentence of the bignum documentation. Perhaps you failed to go on to read the second sentence.

This makes bignum.pm a cute demonstration that can be convenient for a one-liner. Otherwise, you should just use Math::BigInt and/or Math::BigFloat directly, so you can avoid making every single time you use a numeric constant magically turn into a complex object that infects the results of any computations that touch it.

Most of the time you only even need to call Math::BigInt->new() once, so a wrapper like bignum.pm seems of dubious benefit. bignum.pm should probably come with a big warning like "This transforms simple-looking Perl code into something very different and can easily break code".