First of all, a note to anyone who cares: PHP gettext was approved for hosting on GNU's Savannah. I also asked for it to be included in the GNU project, but I don't expect that to happen for various reasons (first of all, it requires PHP which itself is not a GNU project — though, one might argue that GNU gettext is a GNU project, and the goal for PHP-gettext is to add the same interface where GNU gettext is not available).
So, be sure to report any bugs, or send any patches. If you wish to download latest and greatest version, check PHP-gettext project page, and visit either the Files section for released tarballs, or CVS section for instructions on using CVS to fetch the development version.
I've also made some simple improvements in the code which increases performance in some corner-case situations (like reusing of same translation a number of times), but which won't help in general use-cases. If you've got any suggestions, be sure to send them over.
With the old PHP gettext, a stress test of reading in the same string for translation 10.000 times showed the following result:
real 0m23.302s user 0m20.080s sys 0m0.430s
Adding caching of already read in entries has improved the performance three-fold in this particular example (for 1000 reads, the improvement was two-fold, so this can be noticed only in rare occasions), as the following statistics show:
real 0m8.246s user 0m7.340s sys 0m0.150s
Wow, what an improvement in a case no one is going to come across? Wonderful! :-)