well, since you mustn't do anything using realtime (bdb access) inside of the eventloop, that 'd be complicated.
I'd primarily see it for the following things:
- clients (smtp, IGNET, http notify, RSS, Spamassasin)
- maybe also as I/O layer for usual server services; it would have to hand off the processing to worker threads once its shure everything needed is read.The workers write into an outbuffer, which libev would again load off to the client
as for webcit, it migh make sense to start with a similar aproach for the http-frontend; this would enable us to migrate sslg to a long polling mechanism next to without cost. But, since the grammer of the citserver replies are pretty clear it would be easy to have the citserver traffic handed off to libev again; just recreating the stack for the application might be a little more different ;-)