A bit of an init/clear gremlin in integer-expt,

        * numbers.c (scm_integer_expt): Don't mpz_init after scm_i_clonebig or
        scm_i_mkbig, since they do so already.  Don't mpz_clear a bignum SCM,
        since gc does this.

The bad cases are a bignum exponent and a flonum exponent bigger than
a fixnum, so

	(integer-expt 0.5 (ash 1 128))
	(gc) (gc)

and

	(integer-expt 0.5 (exact->inexact (ash 1 128)))
	(gc) (gc)

Run under your favourite malloc debugger to see the duplicate
mpz_clear free when the gc gets around to releasing z_i2.