Ah, no, the Myers algorithm is also known as the "Sortest Edit Script" and your algo does not follow the shortest path or produce a script (just -1 for mismatches).

More specifically, it's not just something at the end appearing at the beginning causing skipping like others have pointed out. The skipping can happen in the middle. Consider the two sequences A B B C and B C A. Your algo will match the first Bs and then the Cs and skip the common B C together.

Even if you could somehow make it work, it would be O(N) but note that with all of the split() and map() and array indexing, it starts to be more like O(N+NX) where X is that potentially non-insignificant overhead.

Don't let me pooh-pooh you from trying but if you do give-in and use the Myers' algo, here's my C diff implementation:

  http://www.ioplex.com/~miallen/libmba/dl/src/diff.c

This is actually pretty compact as diff implementations go and would probably be a reference for implementing diff in JS.