Actually, I guess trans-sid should be implemented as
output-buffering filter, just as gz-compression. Otherwise
what you are trying to do would never work.

I've been thinking about this a bit more.  I'm not sure if adding the ability to obtain the correct length of the buffered output in this case is more important than retailing the ability to save a non-user-specific version of the output.  

"Fixing" my problem would mean that a page caching system that writes plain .html files from database using the ob_ functions would be broken.

Would it be possible to add a function that performs the same function as trans-sid, but at a script level?  With this in place, I could turn off trans-sid and manipulate the output buffer directly.

A generic ob_filter() facility could come in handy for other things too, such as highlighing search terms in a page request.  Especially if it had a switch to filter the HTML, the marked up text, or both.

a) the ob_filter() part is not needed, because output
buffers are nestable
b) would be fine to have ob_transsid or similar, much more
flexible

moving to feature requests

Trans SID is becoming as a legacy feature now. I suppose nobody is going to 
implement this.