Should it also flag multiple votes of one person by another in a short time? Either upvotes or downvotes, since both are equally suspect.
Up votes should have a significantly higher bar, I think. Here's why:
Truly terrible posters tend to be easily recognizable as such, and as such should have no trouble acquiring down votes from myriad sources; serially down voting a bunch of things already at low score isn't adding much value.
A single person serially down-voting someone who is a merely mediocre-to-poor poster is more likely to be a grudge; there's no shortage of mediocre posts to vote down if one is inclined to enforcing high standards, so going after one person is frivolous at best.
On the other hand, there are plenty of excellent posters who focus on niche topics; there may be a limited number of people who can appreciate their posts, and it's easy in a bout of enthusiasm to go "hey, that was neat, I wonder what else this person has posted?" and end up voting them up repeatedly for wholly correct reasons. Perhaps I'm biased (spending most of my time in niche tags) but I care more about rewarding high-quality contributors than catching the low-level abuses this could be mistaken for.
I would imagine that egregious cases, e.g. multiple sock-puppets active over an extended period of time voting up the puppeteer, are rather easy to distinguish from borderline cases.
One negative is that so far community flags are visible to 10k users; I think this particular one should be moderator-only.
Yes, for the simple reason that most people have some idea of "fairness" that they tend to apply without thinking. Publicising a list of serial voting targets is very likely to result in a steady stream of pointless "compensation" votes, most of which would probably be picked up as suspicious as well. Hooray for feedback loops! Insert obvious "stack overflow" joke here, etc. No, I don't think 10k users in general can be expected to know better, unfortunately.
The threshold is also tricky to deal with; should it only take into account a single day? Should 30 votes over two days be flagged? Should 15 votes in one day? What about 10?
A much better heuristic would probably be to look at the sequence of pages the suspicious voter visited, if that were tracked (which it probably isn't, at least not in a useful way for this purpose). Serial voting requires finding things to vote; there are lots of ways to do that, and almost none of them involve looking through lists of questions or searching based on a topic.
Of course, any such heuristics should probably be placed in the same neither-confirm-nor-deny category as the quality filter and such, to prevent gaming (I can think of at least three or four easy-to-implement heuristics off the top of my head, but the less said the better).