I've spent a ton of time in the past week on the 157/154/155 triumvirate in my private scheme implementation.    As of tonight, I've had the last of my eurekas to solve all the space blow-up problems, and have clean interfaces which pass all tests now.

For reference, I copied all the bounded space tests from the (srfi 45) document and run those automatically.

Maybe you already have this all sorted out, but I thought I'd chime rather than wait for your "final" code updates and then raise more issues.

Referencing the "srfi 157 version" of the (srfi 155) sample code...

Using a parameter for current-forcing-extent is not ideal.   It's clear as you already know, that if parameters are based on, say, dynamic-wind, the bounded space tests will fail.   So, presumably, you're depending on parameters being based on continuation marks.  In that case, what you effectively have are two marks getting set in succession -- never otherwise apart.  For  my own code, I simplified this by simply cons'ing c and forcing-extent together into one object and one mark setting.    By doing this, I eliminated any dependency on how parameterize works.

Next up -- setting that continuation mark still blows up the space tests -- because delay captures the current extent, which includes this mark, so each pass of, say, (delay-force (loop)) as the simplest example --  keeps capturing a deeper and deeper list of marks.

I realized that when delay captures the current extent, and when force captures the current extent (for forcing-extent), any values associated with key in those captured extents will never be used, nor can they be observed outside (srfi 155) since all the (srfi 157) interfaces requires that you know the keys you're interested in looking at.    So, I added another API function to (srfi 154) extent-remove-marks which filters a given extent to remove all marks associated with a given key, and used this in those two places where (current-dynamic-extent) was called.

Boom.   All tests pass now, including bounded space tests.