Obviously not the most accurate way to do a benchmark, but just throwing this out there:
# Parrot with it's compiled byte-code
fib(28) = 317811 1.11825299263s
parrot -c fib.pbc 1.12s user 0.01s system 99% cpu 1.130 total
# Parrot
fib(28) = 317811 1.1524600982666s
parrot fib.pir 1.15s user 0.01s system 99% cpu 1.165 total
# Mono C#
fib(28) = 317811
Elapsed: 00:00:00.0094437
mono fib.exe 0.02s user 0.01s system 98% cpu 0.030 total
# Python 2.7
fib(28) = 317811
python2 fib.py 0.36s user 0.02s system 99% cpu 0.379 total
# Ruby 1.9.2p0
fib(28) = 317811
ruby fib.rb 0.13s user 0.00s system 97% cpu 0.136 total
# Perl 5.12.1
fib(28) = 317811
perl fib.pl 0.56s user 0.01s system 99% cpu 0.572 total
In Parrot's defense, they have been putting more work lately into optimizations (ever since Rakudo started to become moderately popular and performance issues came up.) Also keep in mind that parrot hasn't really been at a "stable" version for nearly as long as any of the others this "benchmark" was against. Also, IIRC, the philosophy behind Parrot tends to be, "write the code correctly, then optimize, for its easier to optimize code that's correct to begin with then to try to optimize code that's already broken." Or at least something to that effect.
Anyways, I find Parrot to be an interesting project, that I seriously hope can continue improving so the FOSS community can have a good, general purpose VM to work with (Sure the Mono CLI should be safe since it's an implementation of an open standard, but Parrot doesn't have to worry about even that since it's an original implementation.)