I know next to nothing about Rails, so I'm sorry if these are questions with obvious answers. I just think the decision to process body responses on a per-line basis seems odd.
What's frustrating is that apparently there wasn't any performance testing of the 2.3 release that ought to have found this problem. It's been in the code base for about a year at this point.
My guess as to the reasoning here is so that large files can be served without blocking while the whole thing is written to the socket. Granted, that's just my gut instinct looking at that code, maybe it is just careless...
In the article the author mentions it has to do with serving static files, which is something that nobody should be using Rails to do anyway. I'm not surprised nobody found it until now.
It's true, that the problem only becomes really glaring when the pages being served are quite large (this example was 1300 lines of HTML), but I still assert that good performance regression testing ought to have popped this up. With 100 lines of output, which is pretty reasonable, you'd still be looking at an extra 10ms on each request.
http://is.gd/xP8Y (github link)
http://github.com/josh/rack/tree/master :
* NOTE: String bodies break in 1.9, use an Array consisting of a single String instead.
p.s. this is definitely worth a 2.3.3 tag and release