CMU/CS implemented CSAV and CRET on the 11/40e (6th edition++).  IMO - it was a mixed bag.  I've forgotten the statistics, it was a few ticks fasters for saving the registers, which was good.   The primary thing it did was save a small amount of address space on the non-I/D 40  - which was helpful for applications as they got larger and were running out of address space...... But it was PITA in practice because it meant that binaries compiled on IUS and SUS (the UNIX two systems in CS), would not run anywhere else on campus which were mostly 11/34 or later 11/34A's and it a networked environment (which we were just starting to create) the seams tended to show a little more than we would have liked.

IIRC I eventually got a emulator working at Mellon Institute and EE systems so they binaries would not core dump, but we just recompiled we ran into them. 

The real solution to address space issue was when DEC released the 11/44 which was separate I/D (11/70 class) and the thunk work that we did in 2.9BSD in trying to move 4.1/4.2 code to the 11.

FYI:  Danny Klein and I should have the compiler somewhere (and the microcode) is I ever get my CMU archives off tape.  It's on my list of things to look for.

Clem