Yes, compiled C code is much faster than perl for this task, I'm afraid. I have a C version that runs over 100 times faster than the perl, so it completes in a fraction of a second.
Math::BigInt::Lite should help to address this problem, but the normal Perl overhead will always remain. Unless you use large numbers, where the most time will be spent in the GMP/Pari etc library, Perl will be always much slower, I am afraid.
I see Math::BigIntFast is available in PPM form. Someone who was feeling ambitious could probably make that work, but the rational number support would have to be written.
Math::BigIntFast is just a *very* thin (read: empty) wrapper around Bit::Vector. Besides that Bit::Vector is slower than GMP and Pari, it also has the problem that it only works with "fixed" sized numbers, e.g. you need to determine what the max. will be in front and than use that for all numbers - or do costly resizing all the time.
And if you make it really so that it works with numbers of any size in mixed ways, you will just re-invent Math::BigInt::BitVector ;)
PS: Next time please sent me bug reports or feature requests, it is quite embarrassing reading about things that don't work on some random website you stumble over with google. I could have easily missed it for months or years. Thanx in advance!