For step-over-syscall.exp, we have normally:
...
$ for n in $(seq 1 10); do ./test.sh; done 2>&1 | egrep "expected passes|unresolved testcases"
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
# of expected passes            95
...

But when run in conjunction with:
...
$ stress -c $(($(cat /proc/cpuinfo | grep -c "^processor") + 1))
...
we get:
...
$ for n in $(seq 1 10); do ./test.sh; done 2>&1 | egrep "expected passes|unresolved testcases"
# of expected passes            90
# of expected passes            90
# of expected passes            90
# of expected passes            95
# of expected passes            90
# of expected passes            90
# of expected passes            90
# of expected passes            90
# of expected passes            95
# of expected passes            90
...

We run into:
...
FAIL: gdb.base/step-over-syscall.exp: clone: displaced=off: single step over clone
...

More specifically:
...
(gdb) PASS: gdb.base/step-over-syscall.exp: clone: displaced=off: continue to syscall insn clone
set displaced-stepping off^M
(gdb) PASS: gdb.base/step-over-syscall.exp: clone: displaced=off: set displaced-stepping off
stepi^M
[New LWP 21290]^M
Couldn't get registers: No such process.^M
(gdb) FAIL: gdb.base/step-over-syscall.exp: clone: displaced=off: single step over clone
...