Setting *gamma-imag* to .1d0 (so the region where we use the continued fraction is enlarged), improves the result to

4.93411631550489e+9 %i + 1.911933769523828e+9

Wolfram/Alpha gives

4.9341163155048952691479351508925718965852838271239234... × 10^9 i +

1.9119337695238284027493471541730604272204040299891632... × 10^9

Essentially full accuracy.