Um:), by what criteria ... much better?

If you run the code from your post, but uncomment the last line, you'll see that your algorithm leaves a single duplicate at the end of the array.

With the sample data in your post, this turns out to be the 989. Which is strange as this isn't a duplicate in the input array. That is probably easily fixed though. I had a similar problem with the first cut of my splice version.

However, your algorithm is essentially the same as utils above at Re: Re: Removing Duplicates from Array Passed by Reference and as I showed in my reply at Re: Re: Re: Removing Duplicates from Array Passed by Reference, splice wins against the algorithm because splice doesn't have to copy anything, as it simply unlinks the discarded elements from the linked list that implements perls arrays. I added your version to my benchmark and ignoring the error, splice beats it easily.

Here are the results

             Rate util___sml skeeve_sml splice_sml
util___sml 1368/s         --        -1%       -11%
skeeve_sml 1387/s         1%         --       -10%
splice_sml 1536/s        12%        11%         --


             Rate util___big skeeve_big splice_big
util___big 17.1/s         --       -12%       -24%
skeeve_big 19.4/s        13%         --       -13%
splice_big 22.4/s        31%        15%         --
[download]

As you can see, your implementation beat utils, but maybe once you've fixed the error that may no longer be true. Yell if you want the Benchmark code.