OK, I have now tried with a recompiled maxima.
As I tried to say, the problem mentioned above occurs with maxima 5.41.0 and sbcl 1.4.0, which are the current versions from EPEL--but it seems this maxima doesn't play ball with STACK (it shows the graphs but doesn't differentiate), so I downgraded to maxima 5.39.0 (by "adapting" the spec file from the EPEL source). Now works fine with stack, but still doesn't generate an optimised maxima. I should have trusted Chris's instructions on the maxima version to use!
I further tried two more version of sbcl (since the error seems to occur in its stack handling): a newer 1.4.1 and an older 1.3.18. No luck, the issue persists. I tried building gcl, but the build fails, so that appears not to be an option as a lisp replacement--and I am loath to do anything that can not be done as a build of an rpm, since I am analysing this for a maintainer.
Since maxima 5.38.1 was mentioned, I also tried that. Same issue.
Clearly I can try all allowed maxima's with many versions of sbcl, and avoid the infinite recursion, but I am running out of time: there are serious issues with a RHEL7/Centos7/SL7 installation! Anyone got any other suggestions?
Niels