Michael F.,
In considering a port to mod_(d)tcl, you have to confront the same major issues of moving to FastCGI: separating AOLServer idioms (esp. those not covered by nstcl) from the OpenACS source code. Once you've removed them and are in the land of pure TCL, you're not looking at a lot of work to go with mod_tcl. I'm still getting a feel for what the big crucial AOLServer dependencies are (hence my post), so while I don't think it'd be a huge amount of work, I also don't think it falls under the 'right now' timeframe (as much as I'd like).
Simon, the architectural separation I propose is a very important step when considering the use of a Java/TCL implementation. The more AOLServer-isms sprinked in your code, the more work you'd sign up for such a reimplementation. The decoupling of logic and webserving is much closer to the servlets approach, AFAIK. Also, the KM/webdav implementation I referred to was done using Intraspect/JPython. The combination of scripting and java was pretty slick.
FWIW, here's a writeup I found on fastCGI/TCL.
http://philip.greenspun.com/wtr/cheap-tcl.html
