I'd guess because (1) they're simple, and (2) they're fast enough. Fibonnacci heaps are probably one of the most complicated data structures known to man. Second, O(log N) is not that bad, so there's relatively little to be gained. Even though Fib heaps are O(1) amortized where binary heaps are O(log N), the constants are large enough that especially in Perl, they'll probably cause them to lose for most reasonable values of N. There isn't that much difference between O(log N) and O(1) (or O(log log N) for binomial), so constants can easily outweight the gains. For example:
log N log log N 1
256 8 3 1
1K 10 3.3 1
1M 20 4.3 1
So if inserting an item in a fibonnacci heap is 10 times as complicated as inserting one in a binary heap (not entirely unlikely), the binary heap is faster for 1000 elements.