What is the crowd's opinion of Neko?

From 3.5 billion comments

Of course, but that doesn't explain the rather disappointing results of efforts such as PyPy or Rubinius. If there where so much low-hanging fruits as you say, they should have demonstrated reliable improvements quickly.

It turns out that while they get very good improvements (around 10x) for tight packed numeric manipulations that don't use much kind of abstraction, they're still between "2x faster" and "1.5x slower" in many cases, with possibly memory usage issues etc.
To be fair, it should be noted that the advance of these projects have also been impeded by the various constraint from the languages FFI. Those FFI breaks a lot of the language encapsulation and pose hard constraint on the value representations for example, which impose non-optimal implementation choices. This is not specific to dynamic languages, but still the "our own language is terribly slow, so all performance-hungry operations should be implemented in C through the FFI" mindset is the cause of the abundance of FFI-relying code out there.

That Lua was able to outperform those very quickly, reliably, and with even less manpower (afaik. LuaJIT is mostly the work of one guy) is telling. Even without its JIT implementation, Lua is known to be a well-designed language with a very reasonable implementation (see Lua vs. Neko virtual machines for a very respectful comparison by a competitor), and in my opinion the performance results are only a confirmation of this good work.

This is a kind of moral tale. Do your homework, boy, learn about the state of the art before reinventing your own language, and take care to design something clean and well-specified. If you don't, you'll grow weak, and that will be a hindrance forever.

It's been down for about two or three days now I think, and they are aware of it. There's a GitHub issue open for it here.

You can access the site and downloads through the Wayback Machine. It may be an older page, but it's the newest release (they haven't had a new one in a while).

I just found this: http://nekovm.org/doc/mod_neko

Looks easy enough. Still, I wonder if a similar thing exists for nginx since it's supposed to be a faster server.