I assume you meant "running time" and not "compilation time," since most (all?) of the languages here are interpreted. Regardless, though, I don't think that's a relevant metric. Katas are primarily about writing clean, readable code -- a human metric that has little to do with compilation time or running time.

Another view could be ranking according to test complition time.

I like the idea of recusrsive defintely, mine was heavier than this but check this out...

function solution(l){
return l.map(function(x,i){return l[--i]==x-1&&l[i+2]==x+1?'':x;}).join().replace(/,(,)+/g,'-');
}

I was blown away.

This is much easier to read than your solution.

Good to know, and those are all good ideas. One other thought is to force people (or allow people to earn a little extra honor) by looking at other solutions and deciding to upvote or not to upvote. So you might show one solution per screen with two big buttons for each of those actions, and people would be forced or encouraged to look at some number of other solutions to the problem they just finished before moving on to their next kata. The number would probably need to small if it is required, because forward momentum/addiction is clearly important.

One other idea along these lines (and following HN) is to not show the number upvotes, to eliminate bandwagon bias. Or at the very least, not show the number of upvotes until you had yourself explicitly pressed "Upvote" or "No upvote".

Is there any randomness to the way solutions are displayed?
If not, it means there is a huge bias towards existing upvoted solutions which essentially prevents new, better solutions from ever climbing the ranks.
The system should work similar to Hackernews comments, where new comments start at the top, and then using an algorithm based on avg upvotes per view, start falling back down, sometimes very quickly, and are replaced by the already upvoted solutions.